►
From YouTube: Kubernetes SIG Node 20180904
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
So
essentially,
the
way
the
way
it
operates
is
when
an
empty
door,
volume
or
potentially
any
other
kind
of
volume
wants
to
wants
to
be
created.
It
requests
the
request.
The
quota
met
the
project
quota
mechanism
to
impose
a
quota,
if
that's
possible,
if,
if
it
wishes
to
impose
a
wishes,
not
to
impose
a
quota
but
just
to
use
it
for.
A
Determining
the
amount
of
storage
it
can
do
that
by
simply
requesting
a
very
large,
a
very
large
quota
that
can't
be
exceeded,
say
in
64
max,
for
example,
the
project
quota
mechanism
will
identify
an
available
project
ID,
both
by
looking
at
what
project
IDs
are
in
use
by
the
quota
mechanism
and
by
looking
at
Etsy,
prog,
ID
and
Etsy
projects,
which
are
the
standard
locations
for
this.
At
present,
this
only
supports
ZF
rather
XFS.
A
The
mechanism
is
extensible,
so
it
can
support
other
types
of
file
systems
with
quotas.
Also
at
present
it
only
supports
empty
durval
Yume's.
It
does
not
support
other
types
of
ephemeral
storage,
in
particular
log-log
directories
and
writable
layers.
There
was
considerable
discussion
last
week
about
the
proper
place
to
manage
that,
because
those
are
really
those
are
really
controlled
by
the
runtime,
not
by
not
by
the
cubelet.
A
A
A
B
I
think
the
signal
13
signal
community
amount
to
this
feature
for
a
while
and,
of
course
the
feature
has
a
dependence
there
on
a
newer
kernel.
So
that's
why
we
didn't
development
before,
but
this
is
kind
of
one
of
those
resource
management
roadmap,
and
so
in
your
proposal,
I
noticed
that
I
proposed
the
true
approach
when
monitoring
this
kind
of
things
I
know
driving
invoke
this
through
the
cooper-
directly
another
one.
A
C
A
C
Storage
around
how
to
handle
ephemeral
volumes
or
sandbox
containers
when
we
care
more
about
security-
and
some
discussion
has
gone
on
towards
you
know-
maybe
actually
there
should
be
a
way
for
the
runtime
to
opt
into
managing
volumes
or
we're
just
you
know
in
all
cases
manage
them
as
well.
Yeah
that.
D
A
D
C
D
Guess
Don
some
of
the
other
topics
that
came
up
during
last
week's
discussion.
Well,
many
of
you
I.
Well
we're
what
to
do
when
quota
is
exceeded,
so
I,
don't
think
we
landed
on
a
great
situation
with
that.
So
if
you
did
delegate
to
the
Container
on
time
to
do
enforcement
as
well
as
accounting,
what
did
we
want?
D
I
guess-
and
you
don't
end
up
with
the
state
where
you
have
a
bunch
of
pods
I
could
potentially
be
stuck
terminating,
so
we
we
could
handle
that
a
variety
of
ways
like
you
could
signal
that
the
container
is
out
of
its
allowed
this
quota,
and
then
we
could
extend
the
node
controller
to
do
some
type
of
more
aggressive
action.
But,
frankly,
makes
me
anxious
so
I
didn't
know
with
the
broader
audience
here.
D
If
people
had
strong
opinions
or
if
we
did
delegate
it
to
the
runtime,
how
how
big
of
enforcement
we
know
we'd
want
or
if
we
would
want
to
values
right,
we'd
have
a
like
a
soft
quota
and
a
hard
quota
value
and
the
soft
value
would
be
well
I.
Don't
have
to
think
through
that,
but
I'd
want
a
way
to
ensure
that,
like
pots,
don't
get
stuck
terminating
while
trying
to
write
something
out.
B
A
So
so
my
concern,
my
concern
is
that
if
the,
if
the
ephemeral
storage
is
on
a
particular,
if
it's
on
a
fast
storage
device,
pod
may
be
able
to
write
an
enormous
amount
of
data
before
the
before
the
cubelet.
It
takes
a
look
at
what's
going
on.
If
it's,
if
it's
an
nvme
and
with
an
interval
of
a
minute,
something
might
be
able
to
write
60
gigabytes
or
there
abouts
before
the
cubelet
catches
on
to
what's
happening.
A
B
So
I
have
the
senior
concern.
Is
the
Derek
just
Nick
the
Nate,
the
naked,
the
the
container
run
high,
which
don't
understand
all
the
policy
or
the
skins,
a
negated?
Either
we
can
exceed
everything
Hannah
to
the
continent
and
all
we
delegate
only
partial
to
them
and
inside
a
hard
limit
there
and
if
the
my
housing,
it
is
if
the
Wainwright
cannot
finish
and
sometimes
may
even
cost
based.
My
old
experience
even
me
cause
the
whole
know
the
crusher,
so
so
maybe
kind
of
new
project
quarter
or
ID
don't
have
that
problem.
B
Yes,
issues
exist
today
and
also
so
that's
why
Cuba
night
I
don't
know
the
name
where
we
introduce
the
soft
stress
code
and
how
to
stress
how
to
prevent
out
of
these
complete.
You
are
also
right
because
for
some
faster
storage
we
end
up
like
we
saw
some
customer
and
write
really
fast.
They
do
the
core
down
and
tick
over
the
day's.
Complete
disk,
reverse
work
quickly,
so
and
I
would
still
have
the
problem,
but
that's
the
really
review
how
kisses,
and
also
when
we
really
exam.
B
It
is
within
config
note,
while
because
we
allowed
in
that,
if
you
look
at
those
kind
of
things
we
didn't
come
config
the
node
does
over
a
while,
because
we
didn't
have
faith.
Okay,
how
large
your
code,
amp
can
generate
per
user.
So
some
a
lot
of
things.
We
could
do
better
job,
which
proven
seven
and
my
concern
it
is
in
these
cases
we
basically
puny,
rely
on
another
purely
realized
runtime,
which
is
don't
understand,
Cooper,
night
policy
or
Cuba
night
is
the
policy,
then
we
hand
over.
B
B
B
A
Again,
I
mean
that
that
that
cert,
that
certainly
is
an
option
again
I.
My
recommendation
is
to
use
the
quotas
in
an
enforcing
mode,
perhaps
setting
the
quota
to
be
a
multiple
to
be
some
multiple
of
the
allowed
ephemeral,
storage
I'm
just
I'm
again,
that's
that's
either
way,
that's
a
policy
decision
of
whatever
this
group
decides
it's
a
very
simple
matter
for
me
to
indeed
it
may
not
even
require
any
changed
code
on
my
part,
it's
simply
a
matter
of
how
other
routines
are
called
the
rav
argument
for
the
limit.
D
So
I
guess
Donna
if
we're
talking
about
like
Roberts,
probably
looking
at
trying
to
roll
some
aspects
of
this
feature
into
a
cute
one.
Thirteen
timeframe
right.
So
it
seems
like
there's
general
consensus
on
the
benefits
of
using
project
quota
to
improve
our
monitoring
loop,
and
it
seems
that
there's
broader
debate
or
policy
issues.
We
have
to
work
through
on
how
to
handle
the
mechanism
aspect.
D
I,
don't
know
if
you
do
is
with
you
today,
dawn
or
not,
but
I
think
the
other
topic
that
we
did
want
to
dive.
I,
don't
know
if
we
have
the
materials
today
was.
If
we
do
delegate
to
the
container
runtime,
how
do
we
pass
that
information
down
about
how
to
appropriately
delegate
and
so
I,
don't
know
Robert?
If
you
had
updates
in
mind
for
the
CRI
I,
don't
know
if
it
got
updated
in
the
stock
or
not?
If
not,
we
can
do
that
discussion
a
separate
time.
A
Iii
think
I
think
it
might
be
best
to
have
that
discussion
have
that
discussion
separately,
okay,
but
in
any
I
mean
if
we're,
if
we're
you,
if
we're
using
it
just
for
monitoring
if
we're
using
it
just
for
monitoring,
then
there's
very
little.
If
then,
the
only
thing
we
have
to
change
and
the
runtime
is
to
make
the
call
into
this
into
this
code,
so
I
guess
yes,
I
guess
we
would
need
to
vendor
this
in
any
event.
A
But
all
we
would
need
to
do
is
change
container
runtime
to
make
a
call
into
this
code
to
attempt
to
set
a
quota
and
to
set
an
unlimited
quota
and
then
just
use
it
for
monitoring.
If
we
want
to
use
it
for
enforcing,
then
we
need
to,
then
we
need
to
tell
the
runtime
how
much
space
should
be,
how
much
ephemeral
space
should
be
allowed,
so
the
runtime
can
make
the
appropriate
size
size
set.
E
Yeah
I
was
asking
if
ext4
would
be
supportive,
I,
read,
I,
read
the
doc
and
was
talking
mainly
about
accept
is
one
of
the
main
reasons
why
we
didn't
start
working
on
this.
Probably
a
year
back,
was
that
the
Kota
tool,
support
for
ext4,
wasn't
mature
enough
and
upstream
colonel
like
didn't,
want
to
be
the
pioneer
here
and
like
he
stopped
eating
too
many
issues
with
beer.
So
is
that,
like
reasonably
mature
at
this
point.
A
So
I've
attempted
to
play
around
with
the
ext
ext
for
project
quotas
and
I've,
been
hitting
I've
been
hitting
up
against
some
issues.
I've
been
unable,
at
least
on
my
laptop
to
may,
to
make
it
accept
project
quota,
but
I
it's
the
EXIF
at
the
XFS.
The
quotas
are,
the
quotas
are
apply,
I
have
a
provider
mechanism
so
that,
when
the
amount
that
when
the
file
system
is
identified,
it
delegates
it
off
to
an
appropriate
appropriate
quota
provider.
A
D
Guess
what
is
the
priority
ordering
for
for
you
this
year?
Obviously
at
least
from
the
Red
Hat
perspective,
we're
running
on
top
of
XFS,
and
we
saw
that
as
the
initial
use
case,
driver
I
am
not
sure
if
you're
saying
like
say
we
analyst
as
a
separate
feature
flag
or
a
feature
gate
like
graduation
of
the
feature
to
be
used
for
monitoring.
We
need
to
support
more
than
one
file
system
type
or
what
type
of
staged
rollout.
Were
you
imagining
there
when
you're
bringing
this
up
so.
E
Like
I
mean
basic
talk
process
was
that,
like
it's
either
XFS
if
it's
like
a
radar
based
distro
or
if
it's
going,
it's
going
to
be
X
before
and
that's
like
predominantly,
but
those
are
basically
prominent
file
systems
that
are
being
used.
So
if
we
introduce
a
feature
that
only
works
on
like
wine
partial
file
system,
then
it's
going
to
cause
lots
of
explain
it
over
and
over,
and
so
it
might
be
better
for
us
to
like
not
graduate.
This
feature
be
on
experimental.
Until
we
have
support
for
both
of
those
major
advices
will
be.
A
B
C
A
C
D
D
E
A
The
quota,
the
quotas,
the
quote
the
quotas-
are
the
quotas-
are
stick
so
in
two
ways.
One
is
that
the
quotas
are
sticky
to
the
file
system,
so
they
persist,
so
they
persist.
The
quotas
persist
over
reboot,
I
haven't
actually
done
that
per
se,
but
I
have
I,
have
stopped
and
restarted
the
cubelet
and
the
quotas
are
certainly
persisting
to.
Is
that
I'm
also
storing
the
project,
IDs
and
Etsy
projects,
and
that's
the
approach
I
do,
which
is
the
Linux
standard
location
or
for
storing
those?
A
E
A
E
I'm
supposed
to
be
a
thing
to
the
mapping:
if,
if
naming
is
one
of
the
ways
the
mapping
is
achieved,
then
that's
an
implementation
detail.
I
just
want
to
understand
like
how
one
would
understand
what
project
IDs
belong
to,
what
pods
and
okay
want
to
go
and
introspect.
What's
going
on
on
a
node
and
what
product
IDs
in
a
given
pod
belong
to
a
login
volume
versus
what
belongs
to
a
regular,
empty,
Belgium
or
like
say
a
scratch
directory,
and
so
on.
Okay,.
A
Again,
that
could
be
that
which
we
have
to.
We
have
to
work
out
the
naming,
but
we
did.
We
do
have
to
work
out
the
appropriate
naming
yeah
at
sea,
so
exci
project
stores,
a
mapping
of
a
quota.
Id
well
quote
a
number
rather
directory
on
if
the
directory,
if,
if
it's
possible,
to
determine
from
the
directory
what
it
is,
then
Etsy
projects
will
have
that
information
and
the
mapping
will
be
visible
and
if
you
use
XFS
quota
or
you
use,
the
other
standard
quota
commands
you'll
get
that
mapping
that
way.
A
B
So
maybe
then
we
can
cap,
we
kind
of
agree
in
the
signaled
and
in
the
one
got
a
certain
time
frame
and
we
want
to
work
on
the
project
code
and
I.
Can
we
start
from
the
monitoring
and
do
doing
that
development?
And
then
we
look
into
the
implementing
tail
and
also
the
policy
detail
and
how
we
are
going
to
do.
The
enforcement
and
all
those
kind
of
things.
C
D
D
Have
folks
who
are
established
in
the
community
help
Shepherd
that
feature
and
I
think
I
was
trying
to
think
about
a
way
that
we
can
organize
this
across
the
whole
community.
Like
plenty
of
folks
come
forward
and
say,
I
would
like
to
see
particular
feature
in
kubernetes
and
often
we
we
ask
that
those
interested
parties
like
demonstrate
that
they
want
to
be
around
as
long-term
maintainer
Ziff,
not
just
their
one
feature,
but
to
help
guide
the
node
feature
for
awhile,
and
so
what
I
was
trying
to
figure
out
a
way?
D
D
These
are
my
feature
interest
areas
and
if
we
could
look
towards
whether
that's
113
or
114
planning
trying
to
figure
out
how
we
pair
up
those
folks
with
the
right,
reviewers
and
assignees,
so
that
we
can
scale
the
stake
a
little
bit
more
rather
than
having
like
private.
You
know
back-channel
conversations
about
how
to
enable
this
just
be
very
transparent
on
it.
D
So
I
wasn't
sure
if,
like
folks
that
were
on
the
call
right
now
like
feel
that
they
are
have
an
interest
or
a
particular
passion
on
getting
a
particular
topic
explored
in
a
future,
keep
really
similar.
How
Robert
is
doing
right
now
in
project
Buda
and
if
they
could
make
themselves
known
and
then,
as
we
look
to
do,
113
planning
like
we
could
see
what
we
have
with
respect
to
reviewers
to
help
mentor
those
folks.
But
I
can't
be
the
only
one
that
thinks
that
this
would
be
a
nice
area
for
us
to
grow.
D
B
One
to
join
community
not
just
need
me
to
the
signal
and
they
don't
know
how
to
participate.
They
also
don't
know
how
to
so.
There's
the
couple
issues
and
there
is
currently
the
non-structured
on
and
process
they
don't
know
how
they
grow
themself.
So
some
people
even
mentioned
to
some
people.
If
some
some
people
found
startup,
they
express
their
confusing
and
they
say
the
they
say
the
Fiona.
B
They
need
a
big
company
back
octane,
which
is
not
true,
because
this
is
the
open
source
project
and
but
the
problem
it
is
but
others
hand
I,
feel
they're
feeling
it
is
true,
because
we
need
the
mentor
just
not
be
with
you.
You
miss
an
ecumenical
process
and
grow
those
people,
and
so
I.
Don't
I
can't
I
have
some
sorts
that
I
don't
have
the
really
concrete
idea.
How
to
do
this.
One
thing
I
think
I
can
propose
like
though
we
could
have
some
nectar
forum
and
or
whatever
people
can
register
and
say.
B
B
When
people
join
the
new
team,
they
all
have
the
team,
like
the
mentor
on
their
team
and
also
mentor
cross
team,
to
understand
the
community
at
least
or
not
if
the
Google
culture
needs
or
something
like
that
with
which
is
really
efficient,
I
think
if
there's
the
other,
if
there's
no
other
proposal,
I
will
propose.
We
start
from
there.
If
there's
the
some
other
good
proposal
and
we
could
discuss
in
the
community
yeah.
D
So
I
guess
I
was
just
I
think
that
that
sounds
like
a
great
start.
The
other
thing
generally
was
just
like
if
we
could
have
promised
abla
SH
like
a
budget
for
this
Aegon,
just
how
much
time
we
want
to
Shepherd
towards
like
growing,
crying
folks,
because
I
think
everyone
recognizes
that
we're
reviewer
constraint
and
so
then
there's
like
a
competition
of
ideas
to
get
your
idea
accepted
and
stuff.
D
D
B
B
Yes,
yes,
so
I
just
want
to
clarify
early
I've
just
proposed
that
the
Google
existing
process,
but
it
is
not
meaning
like
that.
Okay
same
company
provided
is
I,
just
think
about
a
the
process.
We
can
apply
to
the
signal
community.
So
then
we
could
have
like
the
people
found
the
same
area
and
you
interesting
scenario,
and
then
we
have
the
reviewer
existing
Liverpool,
so
we
can
find
people
to
mentor
and
at
the
same
time,
since
you
are
new
people,
join
kubernetes
project
and
also
like
the
signal
community.