►
Description
Kubernetes Storage Special-Interest-Group (SIG) Volume Populator Meeting - 02 February 2021
Meeting Notes/Agenda: -
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
C
B
C
It
is,
I
see
okay,
so
hey.
This
is
the
the
volume
populators
meeting
for
the
kubernetes
sig
storage
group.
Today
I
wanted
to
review
the
cap.
I've
I've
updated
the
cup
over
the
last
week
with
all
the
changes
based
on
the
prototyping
work
from
last
year
in
november
and
december,
so
the
the
current
version
of
the
cap
reflects
what
we
actually
want
to
do
for
120..
C
I
wanted
to
thank
shing
for
doing
the
first
review
of
it.
A
lot
of
it
was
like
typos
and
stuff,
but
there
were
two
specific
things
that
well
I
don't
know
if
we
should,
should
we
go
over
the
cap
first
for
those
that
are
not
familiar
with
it,
or
should
we
jump
right
to
the
the
problem
areas
that
need
discussion.
C
So
shane,
I
know
I
know
you've
read
it
yeah
elliot
and
sean.
Do
you
want
to
go
over
the
whole
thing
or
just
discuss
the
problem
areas.
C
So
so
the
the
last
version
of
it
that
existed
for
120
referred
to
a
validating
web
hook.
That
would
reject
pvcs
that
had
invalid
data
sources,
and
so
basically,
I
removed
all
that
language
replaced
it
with
the
new
language
about
this
new.
What
we're
calling
the
data
source
validator
controller-
and
I
just
explained
that
you
know
instead
of
rejecting
object
creation.
It
just
sends
events
and
the
other.
Let
me
see
the
other
major
change.
C
A
C
Well,
I
mean
this
is
this
is
how
this
is,
how
go
land
shows.
A
C
Yeah
yeah,
I
I
like
this.
This
markdown
editor
that
they
have
in
go
land,
it's
nice,
so
the
know
the
big
changes,
I
would
say,
were
in
the
proposal
itself.
So
again
I
explained
here
it's
a
controller,
not
a
web
hook,
and
then
I
added
a
section
where
I
go
into
detail
about.
C
There's
gonna
be
four
different
sources
of
feedback
that
the
end
user
will
get
about
his
pvcs.
So
I
want
to
make
it
clear
that
the
goal
of
all
of
this
is
so
that
the
end
user
knows
what's
going
on
with
his
pvcs.
C
Ultimately,
machines
are
just
going
to
sit
around
waiting
for
stuff
to
bind
right
that
that's
that's
all
they
care
about
is.
Did
the
pvc
get
bound?
Can
I
use
it?
Can
my
pod
start
up
all
of
these.
All
of
these
events
are
for
people
who
are
trying
to
debug
white
something
went
wrong
or
for
an
impatient
user
who
just
wants
to
know
what
the
system
is
up
to.
C
So
I
went
into
detail
on
all
of
the
feedback
that
users
can
expect
to
get
after
this.
After
this
cap
is
completed
and
I'm
still
punting
on
exactly
how
populaters
work
and
I'm
saying
that
will
be
a
separate
kep.
Of
course,
we
have
discussed
how
populated
will
actually
work,
but
but
that
I
don't
think,
is
that
that's
sort
of
out
of
scope
for
the
the
main
kubernetes
api.
C
All
we
need
is
to
get
the
the
data
source,
any
any
data
source
feature
gate
moved
to
beta
in
order
to
move
things
forward
in
kubernetes
proper
and
then,
of
course,
all
the
actual
population
stuff
is
going
to
be
out
of
tree,
whether
we
actually
need
a
cap
for
the
out-of-tree
stuff
or
not.
I
don't
know
I'm
happy
to
write
another
kept
for
you
know
how
populaters
work,
but
I
wanted
to
point
out
that
this
this
document
doesn't
explain
the
mechanism
of
population.
C
It
just
explains
that
that
the
expected
result
is.
It
looks
like
any
other
data
source,
or
you
know
it
looks
like
exactly
what
happens
if
you
specify
a
data
source
of
like
a
snapshot
or
another
pvc,
it
just
gets
created
and
binds
and
you're
ready
to
go.
This
section
is
all
the
same.
C
C
Anything
here
I
had
to
update
the
test
plan
to
explain
that
what
we're
looking
for
now
is
whether
events
get
generated
or
not.
When
I
missed
some
periods
here,
pop
those
in
so
so
shane.
I
I
instead
of
saying
ignores
the
pvc
I
said:
causes
no
events,
that's
more
neutral
language,
because
I
agree
that
saying
ignores
sounds
negative
yeah.
It's
a.
A
Neutral
language
here
main
thing
that
I
see
is
now.
I
just
realized
one
thing
since
you
are
not
going
to
use
the
sidecar
right
so,
which
means
we
don't
have
that
matrix
mechanism
like
anymore.
So
I
don't
know
if
this
is
a
problem
for
the
no,
the.
C
A
C
It
for
every
pvc
under
the
responsibility
of
side
driver
the
provisioner
sidecar
for
that
pvc
will
emit
an
event
explaining
the
sidecar
is
not
taking
action
on
the
ppc.
Assuming
the
data
source
is
something
that
a
popular
responsible
for
and
not
a
pvc
or
volume
snapshot
which
the
sidecar
is
responsible
for.
So
right.
A
But
you
are,
I
mean
the
country,
the
new
controller,
I
guess
the
I
guess.
The
thing
is,
the
new
controller
will
not
have
any
metrics
right,
so
I
think
that's
something
normally
that
is
required,
so
I
don't
know
like
for
for
us
right
for
snapchatter.
A
A
Oh
so
well
so
look
the
matrix
basically
like
for
like,
if
you
have
some
controller
or
something
in
your
new
feature.
Supposedly
it
should
have
some
metrics
right
so.
C
C
Yes,
yes,
so
absolutely
I
want
to
talk
about
metrics,
that's
on
the
agenda,
okay,
but
but
there
was
one
other
issue
that
you
pointed
to
that
I
wanted
to
get
to.
Okay.
A
Sure
that's
fine
yeah,
I
just
I
I
just
to
me.
That
seems
to
be
the
the
main
thing
that
I
just
realized.
Last
night,
when
I
was
reading
this,
I
said.
Oh,
this
seems
to
be
new
because
I
think
previously
it
was.
We
were
talking
about
a
maybe
we're
talking
a
set
car
or
something.
So
I
thought
side
car
would
work,
but
now
we
have
a
new
controller,
does
not
have
matrix.
So
I
don't
met
sorry
matrix
and
I
don't
know
what
is
the
you
know
requirement
for
the
this
new?
A
C
There's
there's
a
ton
of
stuff
in
in
the
feature
enable
in
the
product
readiness
review
questionnaire.
A
Yeah,
so
that's
why
I
don't
know
if
they,
if
we
don't
have
that
in
for
beta,
I'm
not
sure
if
that's
a
problem,
so
I
just
I
just
wanted
to
bring
this
up.
I'm
not
sure
this
is
the
first
time,
because
I
know
that
for
snapchatter
for
snapshot
controller.
Well,
I
think
cycle
has
that.
But
the
snapshot
controller
didn't
add
that
on
to
before
ga.
But
I
don't
know
if
you
know
now
they
have
a
different
requirement
or
what,
for
this.
C
Yeah-
and
I
did
get
a
comment
from
one
of
the
one
of
the
upstream
people
about
needing
to
create
a
different
file.
Maybe
I
need
to
move
all
the
prr
stuff
into
a
different
file.
Oh.
A
No,
no!
It's
okay!
I
can
show
you
what
what
they
mean.
No
you
don't
have
to
put
that
in.
This
is
fine.
This
is
there
here.
You
just
need
to
just
create
another
small
tiny
file
like
three
line
file
that
include
the
name
of
your
of
this
pr
another's
pr
this.
What
is
that
the
issue?
Number
okay,
like
cap
number,
and
also
who
reviewed
this,
who
approved
this?
That's
it.
I
can
show
you
what
what.
A
I
will
just
send
you
just
like
after
this
I
will
send
you
like
what
I
have
for
the
voting
house.
I
added
that
already.
Is
there
just
a
tiny
little
file.
C
So
so
the
specific
contents
of
the
prr
that
I
wanted
to
talk
about
first
was
the
talking
about
disablement
rollback.
So
if
someone
does
install
this
feature
and
they
decide,
they
don't
like
the
way
it
works
and
they
want
to
disable
the
feature
gate
that
the
pr
the
the
product,
the
production
readiness
review,
asks
you
to
explain
how
that's
going
to
go.
So
my
understanding
is,
if
you
flip
the
feature
gate
off,
and
I
guess
that
requires
bouncing
your
kubernetes
services.
C
What
will
happen
is
that
the
the
old
code
in
the
admission
controller
will
be
enabled
again
and
then
every
time
a
pvc
gets
loaded
from
from
ncd
the
admission
controller
code
will
run
on
it
and
it
will
drop
the
data
source
fields.
That
is
my
understanding
of
what
happens.
C
So
I
wanted
to
ask
if
other
people
had
the
same
understanding
that
if,
if
you
turn
the
feature
gate
off,
so
the
old
admission
controller
logic
is
re-enabled
and-
and
you
have
stuff
in
your
xcd
database,
where
you
have
data
sources
set
to
something
that
doesn't
pass
the
old
admission
controller
logic.
Will
it
just
drop
it
as
if
it
was
never
there?
D
C
C
So
it's
just
historical
information
at
that
point,
but
if
you
had
any
like
unbound
pvcs
hanging
around
and
then
you
shut
down,
cube
api
server
and
disable
the
feature
gate
and
then
started
it
up,
I
imagine
that
it
would
just
zero
out
those
fields
and
then
you
get
a
bunch
of
empty
pvcs
like
you
would
today,
if
you
had
never
enabled
it
so
I
may
need
to
do
the
experiment
to
determine
exactly
how
this
behaves.
If
nobody
knows,
but
I
was
hoping
somebody
would
know
how
how
that
works.
C
A
C
Are
are
there
any
missing
metrics
that
would
be
useful
to
have
so
yeah?
I
wanted.
I
just
wanted
to
get
some
suggestions
like
what
what
type
of
metrics
might
make
sense
for
this
controller
that
I
can
put
in
here,
because,
because
remember,
the
purpose
of
this
controller
is
just
to
give
you
feedback
on
pvcs
that
look
like
they
might
not
bind
ever.
C
The
actual
populators
will
probably
have
all
kinds
of
metrics
on
what
they're
doing,
because
they're
doing
real
work
to
go
populate
a
volume
and
so
they'll
have
like
operation
counts
and
timers
and
how
long
it
took
and
how
many
failures
and
how
many
successes
there
were.
But
this
thing
is
just
just
you
know
informing
you
when
you
have
suspicious
looking
data
sources,
I
guess
maybe.
D
B
Counter
for
number
of
pvcs
with
data
sources
that
what's
the
right
word
to
use.
A
C
C
D
C
Well,
I'm
just
trying
to
think
about
like
is
there
something
in
the
api
that
we
can
hook
into
to
determine?
If,
if
this
is
the
first
time
we've
we're
noticing
a
pvc
and
complaining
about
it
or
if
it's
just
you
know,
the
pvc
got
updated
and
showed
up
in
the
watch
again
yeah
like
if,
for
for
one
running
instance,
we
could
maintain
a
map
in
memory
of
which
ones
we
had
counted,
but
then,
if
it
ever
restarted,
you'd
lose
your
your
state
right
and
there's
no
place
to
stash
that
state.
C
Your
counter
would
go
back
down
to
zero,
though
right.
If
the
poverty
started.
Well,
I
actually
don't
know
how
the
metric
system
works.
Yeah.
I
guess
that's.
Another
point
is
if,
if
the
metrics
themselves
are
just
in
memory,
data
that
has
to
get
scraped,
then
yeah
a
restart
of
the
pod
would
also
reset
all
those
to
zero.
C
C
So
I'm
happy
to
do
that,
implementation
that
that
will
be
a
little
bit
of
new
code
in
this
controller
which
will
need
to
get
sorted
out
soon.
If
we're
aiming
for
121.
C
yeah.
I
can
do
that
so
that'll,
be
another
action
item
for
me:
is
figure
out
how
to
do
the
metrics
and
get
get
some
at
least
stateless
metrics
in
there,
and
then
we
can
avoid
double
counting
for
at
least
that
invocation
of
the
controller.
So
if,
if
the
same
pvc
comes
in
twice,
we'll
only
count
once
that's
easy
to
do
in
memory,
okay,
so
I've
got
two
action
items
now
and
the-
and
I
don't
know
if
there
was
anything
else
to
cover
in
in
the
cap
itself.
C
We
got
the
oh
and
I
need
to
do
the
that
create
that
other
three
line.
Yaml
file
and
shingle-
help
me
with
that
and
that's
it.
I
guess
I
will
make
those
changes
and
push
another
update
to
my
pr
who,
who
else
do
we
need
to
get
reviews
from
to
get
through
the
the
kep
review
process
and
get
it
merged
for
121.?.
C
C
Okay,
so
so
yeah
after
I
after
I
have
it
in
a
good
shape,
I'll
sod
urine
approver
tim
hawkin
is
a
reserve
approver.
Oh,
I
have
david
eads
for
the
for
the
my
prr
approver
at
least
he
was
last
time
around.
I
don't
know
if
he's
still
doing
that
work,
but
I'll
ping,
him.
C
Okay,
okay
yeah.
He
was
very
friendly
when
I
approached
him
like
three
months
ago,
so
yeah
and
then
we'll
try
to
get
it
reviewed
and
merged
shin.
Can
you
remind
me
what,
when
exactly
is
the
deadline
for
that.
A
Oh,
it's
next
tuesday.
Actually,
it's
coming.
C
Okay,
so
I've
got
seven
seven
days
basically
to
to
wrap
this
up.
I
think
I
think
it'll
be
easy,
because
yeah
the
the
remaining
updates
I
can
do
today,
I
can
get
out
review
requests
and
then
talk
to
anyone
that
needs
convincing
or
to
have
any
discussions.
We
need
to
have
later
this
week.
C
Okay-
okay!
So
that's
that's
it
for
for
this
week,
then
I'm
just
gonna
try
to
get
the
kept
merged
before
the
next
meeting
and
and
yeah
all
right.
C
Do
those
action
items
I
mentioned
no
I'll
write
those
down
after
after
the
meeting
okay.
So
unless
anyone
has
anything
else,
we'll
we'll
end
for
today,
all
right
take
care.
Okay,
thank
you
bye.
Thank
you.