►
From YouTube: Kubernetes SIG Storage 20180118
Description
Kubernetes Storage Special-Interest-Group (SIG) Meeting - 18 January 2018
Meeting Notes/Agenda: https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.aukk6ioq19p7
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
Moderator: Saad Ali (Google)
Chat Log:
09:34:00 From Eric Forgette : Can we add snapshot resource definitions and leave the controller out of tree?
09:36:11 From Luis Pabon : +1 for new goal
09:36:13 From Ardalan Kangarlou : I think that might be a good middle ground, very similar to how external provisioners operate on the same PVCs and PVs but they can be implemented in drastically different ways.
A
A
A
Essentially,
what
that
means
is
you
have
to
create
a
feature,
repo
bug
associated
with
your
feature
and
make
sure
that
it
is
marked
for
with
the
1.10
milestone.
So
this
bug
here
needs
to
be
updated
for
the
red
blocks.
Here
we
need
to
make
sure
that
a
feature
repo
issue
is
created,
otherwise
it
won't
be
tracked
properly
for
this
release.
A
B
C
This
feature
changes
how
bad
works
and
has
been
testing
only
on
a
few
distribution
and
few
kernels.
So
anyone
I
asked
if
anybody
uses
on
propagation,
in
suicide
work
or
for
flex.
Listen.
Mere
feedback
I
need
to
know
how
it
was
tested,
because,
if
out
reasonable
insurance
that
it
works
everywhere,
I
am
afraid.
I
won't
approve
it
for
beta
for
GA,
because
it
will
break
things.
C
D
C
A
B
F
D
A
D
G
Block
volumes
that's
out
there.
In
addition,
Scott's
working
on
the
AWS,
plugin
I'm
gonna
start
working
on
the
GCE,
the
Gluster
plug-in
from
the
CNS
team.
I
heard
they're
ready
for
that
now
to
have
the
proper
volume
mode
into
their
block
provisioner.
So
well,
those
are
the
ones
we'd
hoped
Oh.
In
addition
to
I
scuzzy
has
been
merged,
so
those
are
the
ones
we
had
slated
for
this
release:
yep
Aaron,
who
CNS
I'm,
sorry
Gloucester,
container
native
storage.
They
had
a
block
volume
provisioner
for
Gloucester
prior
to
this
release.
G
H
I
J
J
A
H
J
A
E
H
A
J
A
A
Lynn
I,
don't
know
if
you're
on
the
line
and
if
you
want
to
add
anything
else,
I
think
Lynn
is
in
China
and
it's
the
middle
of
the
night
there.
So
I
don't
expect
him
to
be
here
all
right.
Next
up
is
CSI
volume
plug-in
implementation.
We
are
tracking
the
status
of
this
project
in
the
CSI
implementation
meeting.
You
can
go
there
for
a
detail
of
what
items
we're
tracking
for
this
quarter.
A
Work
has
begun
we're
going
to
attempt
to
move
the
CSI
volume
plug
into
beta
this
quarter
earlier.
We
were
concerned
with
the
amount
of
work
that
was
required,
but
what
we
decided
is
that
we
don't
need
to
move
every
single
feature
that
is
part
of
CSI
to
beta
all
at
once.
We
can
move
the
core
CSI
API
to
beta
and
then
add
on
additional
features
as
alpha
and
move
them
to
beta
separately
or
independently
of
the
core.
So
this
allows
us
to
decouple
from
things
like
block
support
and
things
like
that.
A
A
A
B
B
A
J
J
A
H
So
we
want
to
push
to
move
snapshots
API
to
entry,
but
we
definitely
need
some
like
very
strong,
so
support
and
documents
to
present
to
see
architecture
so
because
we
need
approval
from
them
to
get
it.
Entry
and
I'm
working
on
the
document
and
the
tool
prison
so
that
we
can
present
to
the
Sikh
architecture
and
so
we'll
in
this
meeting.
We'll
talk
a
little
bit
about
something
we
were
to
discuss
and
anyone
interest
who
can
help
us
on
the
tube
yeah.
A
Some
of
this
has
to
do
with
the
fact
that
we're
working
on
imperative
operations
and
trying
to
work
them
into
a
declarative
API.
So
we
need
to
work
on
figuring
out
the
API
around
the
restore
operation.
Another
challenge
is
thinking
about
it
more
holistically.
How
would
higher
level
primitives
like,
for
example,
stateful
sets
utilize?
This
snapshot
and
restore
ability,
specifically
the
restore
what
would
a
restore
operation
for
a
deployment
sell
for
a
stateful
set,
look
like
and
then
finally,
once
we
have
those
design
questions
nailed
down.
There's
the
big
question
about.
A
Well,
you
have
this
thing
working
out
of
tree:
why?
Why
do
you
want
to
bring
it
into
tree
into
the
core
kubernetes
codebase,
as
many
of
you
are
aware,
kubernetes
is
becoming
more
and
more
extensible
and
decoupled
and
there's
a
very
high
bar
for
what
should
be
part
of
the
core
kubernetes
or
what
they're
calling
the
nucleus.
A
Our
argument
largely
has
to
do
with
the
fact
that
it
makes
a
user
experience
a
little
bit
easier
and
that,
having
this
be
part
of
the
core
will
allow
higher
level
primitives
to
be
built
on
top
of
it.
So
that's
the
argument
that
Jing
is
going
to
try
to
push
for
and
then
present
to
the
sig
architecture
group
and
then
we'll
proceed
from
there.
If
we
get
approval,
then
we
can
start
moving
towards
pulling
this
code
entry.
E
Just
just
to
add
thank
you
for
the
explanation.
Uh-Huh
I
think
that
if
our
proposal
is
to
do
entry
before
we
did,
CSI
I
really
agree
with
it,
but
now
that
we
have
done
CSI
and
we
have
three
side,
car
containers
that
are
outside
the
tree
and
we've
got
probably
going
to
have
more
and
this
the
responsibility
of
these
storage
vendors
to
add
its
convenience
right
and
responsible
for
these
outside
components.
I
I,
don't
I'm,
not
really
convinced
that
we
need
to
have
snapshots
in
the
tree
anymore.
E
A
And
that
may
well
be
the
the
path
that
we
go
down.
I
mean
I,
think
the
big
difference
between
what
we
have
with
CSI
and
what
we
have
today
versus
having
an
out
of
tree
snapshot
controller
is
that
the
or
logic
for
understanding
the
existing
operations
like
provisioning,
attaching
mounting
resizing
all
these
things
exist
in
the
core,
kubernetes
and
kubernetes
understands
how
to
do
them.
Natively.
A
The
only
thing
that
is
extensible
through
CSI
and
flex
is
the
ability
to
add
additional
storage
providers
who
support
that
functionality
and
what
we
would
be
doing
if
we
push
to
keep
snapshots
out
of
tree.
Is
that
we're
saying?
Not
only
do
you
add
additional,
you
know
additional
storage
vendor
functionality
out
of
tree.
You
had
you
know
core
storage
functionality
through
an
external
controller
as
well,
which
makes
which
raises
the
bar
for
being
able
to
run
the
storage
system
right.
A
A
yellow,
because,
if
right
now
having
it
out
of
tree,
causes
some
user
visible
friction
so,
for
example,
the
way
that
it's
currently
implemented.
If
you
want
to
do
a
restore
operation,
you
have
to
have
a
PVC
that
is
pointing
to
a
snapshotting
provisioner
right.
You
can't
use
your
standard
provisioner,
because
your
standard
provisioner
doesn't
understand
snapshotting
things
like
that,
but
I
agree
I,
think.
Ultimately
it
will
be
a
very
high
bar
to
get
an
entry
and
will
at
six
storage
or
a
cig
architecture,
make
the
calls.
K
L
K
K
L
L
Think
she
thinks
snapshot
is
well.
You
could
have
different
policies
but
like
who's,
employing
over
stack
sender,
rights
and
snapshots
teacher.
This
is
a
required
feature
from
day
one,
it's
just
a
naturally
part
of
it
right,
so
I
just
couldn't
think
of
it
as
a
something
has
to
be
out
of.
She
hasn't
not
part
of
this
switch
itself
right.
When
you
take
new
creative
watering,
then
you
want
to
take
snapshot
to
protect
it
race.
You
take
some
point
putting
time
copies
of
that.
It's
part
of
that
storage
itself.
It's
different
from
some
other.
L
M
M
A
Absolutely
critical-
and
that
needs
to
happen
sooner
or
later,
whether
it's
in
tree
or
out
of
tree
in
order
to
allow
storage
vendors
to
be
able
to
implement
support
for
snapshotting
I'm,
not
sure
how
that
would
influence
the
decision
of
the
core
controllers
that
run
snapshotting
to
go
in
tree
or
out
of
tree.
I.
Think
those
two
can
happen
independently.
L
That's
for
the
you
know
it's
for
the
drivers
right,
so
it's
the
stores
you
can
face,
but
for
the
API,
its
students
not
sort
of
the
problem,
the
API
right
still,
you
know,
if
you
you
know,
we
of
course
definitely
should
out
in
CSI,
but
if
we
don't
either
in
tree,
your
API
definition
is
still
not
part
of
kuben
ad.
So
so
it's
like
it's
not
a
core
feature.
Yeah.
A
A
A
that's
an
important
point,
because
when
you
have
something
that
is
exposed
as
part
of
the
core
kubernetes
api,
then
you
can
have
other
components
of
kubernetes
depend
and
build
on
top
of
it
right.
So
if,
if
snapshots
API
exists
as
part
of
the
core
kubernetes
api,
then
you
can
start
building
higher
level
primitives
on
top
of
it,
you
could
have
stateful
sets,
for
example,
depend
on
this
behavior
to
build
more
advanced
functionality.
I
think
that's
probably
the
key
argument
for
for
pulling
it
in
tree.
I.
I
Sorry,
the
main
danger
of
having
it
out
of
trees
that
lenders
Kenney
implements
yard
is
how
are
they
alike
and
back
to
the
Lewises
comment
about
the
experience
from
an
engineer's
point
of
view
from
the
users
point
of
view.
This
is
really
about
the
experience
from
the
operators
point
of
view
the
admins
imagine
if
each
vendor
is
going
to
have
their
own
Co.
Are
these
made
a
different
format,
different
structure,
different
workflow
and
now
the
operators
have
to
you
know
you
see:
are
these
in
different
ways
for
different
vendors?
I
A
I
A
Sounds
like
there
is
a
disagreement
about
this.
Let
me
suggest
something
so
this
quarter,
instead
of
making
the
goal
trying
to
pull
this
entry,
let's
modify
the
goal
and
make
it
let's
try
to
remove
the
the
friction
that
exists
in
the
API
today
and
add
CSI
support
for
this
quarter,
and
then
this
next
quarter-
and
hopefully
we
can
do
a
face-to-face
meeting
around
that
time
and
we
can
all
get
in
a
room
and
kind
of
hammer.
This
out
there
yeah.
K
I
K
D
N
A
Is
fun
we
kuber
does
not
necessarily
control
CSI,
so
what
this
would
involve
is
going
to
the
CSI
community,
proposing
adding
snapshots
as
an
extension
to
the
existing
calls
that
are
available.
If
and
when
that
gets
merged,
then
updating
the
external
snapshotting
controller
to
be
able
to
speak
CSI
I
think,
regardless
of
whether
we
go
in
tree
or
out
of
tree,
that
is
going
to
be
critical
for
us.
A
O
O
A
A
L
A
H
O
L
A
L
A
A
A
A
P
So
the
container
identity
working
group
is
working
on
a
proposal
for
her
node
identity,
we're
trying
to
understand
the
timeline
and
the
scope
of
that,
because
the
result
of
that
is
going
to
affect
what
our
design
is
going
to
turn
out
to
be
so
just
working
with
them
and
figuring
out
exactly
what
what
is
happening
and
how
long
it's
gonna,
how
long
it's
gonna
take!
Okay,.
A
A
O
P
Q
A
D
F
A
That
offline
and
let's
keep
moving
down
our
agenda
after
status
updates
mount
propagate
needs
more
intensive
testing.
This
is
an
important
feature.
We
need
your
help
with
it.
Please
visit
the
link
that
Yann
posted
here.
Otherwise
it
may
be
in
jeopardy
for
this
quarter.
Next
up
is
PRS
that
need
attention.
Jing
paste
it
in
about
the
scalability
issue.
Jing,
do
you
want
to
talk
about
this.
H
I
think
I
hit
this
problem
when
working
with
some
customer,
and
we
want
to
have
some
providing
and
collect
some
data
to
see
whether
we
can
handle
I
say
hundreds
of
thousands
of
volumes
per
no,
and
if
anyone
interesting
like
volunteer
this
work,
I
can
work
with
them
to
do
some
profiling
collecting
data
and
see
how
we
can
solve
these
kind
of
issue
and
another
one
is
right
now
for
each
volume
operation
we
have
a
separate
goal
routine
and
we
don't
have
any
limits
on
how
many
core
team
we
can
start.
In
some
cases.
H
A
Okay,
so
if
anybody's
interested
in
helping
look
into
scalability
issues,
please
reach
out
to
Jing
Jing
has
a
lot
on
her
plate,
as
do
many
of
us.
So
this
is
not
prioritized
very
highly
for
us
at
the
moment.
If
this
is
super
critical
for
you,
I
highly
recommend
you
reach
out
to
Jing
and
take
a
look
at
this
issue
and
see
how
you
can
help.
Otherwise
we
will
try
to
get
to
it
as
soon
as
possible,
which
may
be
next
quarter.
H
Ways
to
fix
the
problem
we
had
before
because
for
some
I
say
maybe
half
of
the
worn
plugins
they
cannot
construct
the
bottom
specs
based
on
just
factory
the
past
information
and
the
new
design
is
or
those
for
fighting
that
cannot
support.
There
is
construction
and
though
we
can
do
that,
just
clean
up
clean
the
mounts
they
have
today,
instead
of
restruck
into
a
state,
we
still
want
to
keep
restriction
because
it
gave
us
benefit.
Hope,
let's
say
if
you
clean
out,
fails
right.
This
is
in
the
actual
state.
H
A
R
G
S
S
Protein-Protein
support
chose
one
point:
there
I
got
some
comment
from
yeah
I'm
gonna
share
a
country
updating
speak
so
in
basic.
This
design
dog
explains
how
to
implement
rock-bottom
for
internet
for
Ghana
and
eczema.
So
I
think
there
isn't
a
big
change,
but
point
we
need
to
consider
is.
There
is
a
case
that,
if
probably
and
broke
two
dozen
supports
rock
bottom,
but
user
can
sit,
volume
was
as
a
block
in
that
case
people
on
the
people
she
created,
but
they
can't
be
bound
and
they
stuck
and
pending
state
pending
status.
So
we
need
to.
S
A
H
A
Okay
Nate,
so
it
sounds
like
the
next
steps
here
are.
There
is
going
to
be
meetings
that
are
going
to
be
set
up
in
the
next
within
the
next
week
or
two
to
discuss
whether
we
want
to
move
entry
or
not.
They
will
be
led
by
Jing
and
saying
and
based
on
what
is
decided
in
those
meetings,
we
will
decide
whether
to
proceed
with
entry
or
not
or
focus
on
the
other
items
that
are
left
for
snapshots
next
up
is
postpone,
PV
deletion
if
it's
bound
to
PVC
or.
C
Not
yeah,
it's
a
proposal
open
by
further
Nicorette.
He
calls
himself
on
github,
but
anyway
it's
it
copies.
Basically,
the
structure
of
PVC
protection
controller
and
the
main
message
here
is
that
it's
going
to
be
a
separate
controller,
so
we
will
have
one
controller
for
PVC
protection
and
second
one
for
pv
protection.
C
C
C
A
K
Sir,
so
there
has
been
an
increasing
pressure
from
the
broader
kubernetes
community
for
us
to
provide
status
on
a
regular
basis
and
I.
You
know
they
vest
initially
they
you
know
they
wanted
it
to
be
part
of
the
community.
The
today's
community
called
that's
right
after
this
they're
moving
into
a
separate
call,
I'm,
not
sure
but
I
kind
of
wanted
to
get
feedback
on.
K
K
G
K
Other
teams
do
present
status
and
it's
a
little
haphazardly.
I
mean
like
there
isn't
like
one
specific
format
that
everybody
follows,
but
another
part
of
it,
too,
is
like
we
have
a
lot
of
vocal
people
in
this
community.
I
I
would
need
to
see
it
just
be
like
me
and
start
to
make
sense.
Every
time.
I
think
there's
opportunities
for
other
people.
You
know
yeah.
A
I
think
so
so
from
the
community
meeting
organizers,
a
message
that
we
received
to
the
cig
leads.
They
said
one
of
the
challenges
that
they
had
was
they
had
to
reach
out
to
individual
SIG's,
to
schedule
them
to
come
in
and
talk
at
the
community
meeting,
which
becomes
a
pain
for
them,
and
so
what
they
want
to
do
is
set
up
regular
reoccurring
time
slots
twice
a
quarter
for
a
sig
to
come
in
and
give
an
update.
A
Those
would
be
decided
at
the
beginning
of
the
quarter,
which
timeslots
we
have
and
to
Brad's
point.
You
know
me
and
Brad
can
always
go
in
and
give
an
update,
but
it
might
be
a
good
opportunity
for
folks
in
this
thing
to
get
exposure
to
the
broader
community
and
perhaps
deliver
a
more
diverse
message
than
then.
You
know
with
me:
Oh
Brad
can
deliver
so
for
for
the
upcoming
community
meetings,
where
the
six
storage
gets
slots
to
speak.
G
K
N
A
And
the
other
thing
to
keep
in
mind
is
that
the
community
meeting
always
has
a
slot
for
something
like
fifteen
or
twenty
minutes
in
the
beginning
of
the
meeting
to
do
a
presentation,
a
demo
if
anything,
interesting
or
cool,
there's
a
lot
of
cool
things
that
come
out
of
this
SIG
and,
if
you're
ever
interested
in
broadcasting
it
more
widely.
That
is
an
opportunity
that
is
available
to
you.
You
can
reach
out
to
the
community
meeting
organizers
directly
or
reach
out
to
me
and
Brad,
and
we
can
help
set
that
up
as
well
as
nap.
A
You
can
treat
that
as
an
avenue
to
get
wider
exposure
and
yes,
I,
agree,
Steve,
I
think
what
we'll
do
is
next
time
we
get
a
slot
to
do
an
update,
for
this
sake,
we'll
come
back
to
the
sig
and
we
can
discuss
it
here
and
see
what
folks
want
to
present
and
who's
interested
in
presenting
sounds
good
to
me
all
right,
Erin.
We
have
two
minutes
left
I
apologize
for
that.
Do
you
want
to
talk
about
it
now
quickly
or
do
you
want
to
punch
it
to
the
next
meeting.
G
Oh,
let's
mention
I
sent
out
an
email
earlier
today.
Everyone
got
it
on
list
about
15
different
topics.
We
feel
it
would
be
good.
Onboarding
topics
try
to
make
it
fun
and
interesting.
I
am
looking
for
volunteers
for
each
of
these
topics
to
plan
a
presentation
that
would
be
recorded,
I'm
and
archived
for
future
on
borders.
This
is
to
provide
mentorship
and
knowledge
transfer
within
our
community
of
people.
G
G
Think
we
could
offer
a
meeting
that
people
want,
because
I
honestly
think
that
when
people
have
questions
and
those
are
captured
in
the
recording,
that's
also
helpful
instead
of
people
having
to
go
to
the
muttering
again
and
again,
decisions
so
I
just
set
up
a
time
for
each
one
of
those
to
record
it
and
people
could
come
and
and
have
that
interaction
and
you
schedule
accordingly,
once
a
thing
bothers.
Okay,.
N
A
Awesome
so
next
steps
here
are:
please
take
a
look
at
the
message
that
Aaron
sent
to
the
storage
sig
group,
if
you're
interested
in
volunteering
to
do
a
presentation
please
reply
to
that
reach
out
to
Aaron.
If
you
think
that
there
is
a
topic
that
may
not
be
covered,
especially
if
you're
new
to
the
group
and
there's
something
that
you
really
wish
was
covered,
please
reply
to
that
email
as
well
and
suggest
the
topic.