►
From YouTube: 2017-10-09 Rook Community Meeting
Description
Rook bi-weekly community meeting
B
C
B
D
B
B
B
B
B
D
B
D
D
B
C
Was
one
of
the
bigger
issues
where
we're
doing
a
lot
of
things
on
the
on
the
driver
when
I
say
a
lot
of
things
specifically
mounting
and
formatting
and
then
by
some
suggested,
to
move
that
into
the
agent
now?
The
reason
why
it
was
done
on
the
driver
side
is
because
the
driver
actually
runs
on
the
context
of
the
host
and
the
agent
doesn't
know
ahead
of
time
where
company
is
suspecting
directory
to
be
mounted
and
formatted.
So
we
can
do
a
best-guess
by
you
know
by
by
giving
the
agent
ahead
of
time
hey.
B
B
C
B
Just
made
a
pass
through
all
the
issues
and
it
looks
like
there
is
a
I
think,
there's
like
fifteen
or
twenty
issues
that
are
either
going
to
be
closed
because
of
the
rogue
agent
or
are
somehow
pending
after
after
that.
But
maybe
what
you
can
do
Steve
is
you
know
rebase.
It
will
merge
it
and
then
make
it
pass
through
all
the
issues
assigned
to
you
either
close
them
or
you
know,
label
them
appropriately
or
add
new
ones
coming
out
of
things
that
fell
out
of
the
review.
Sure.
C
B
C
B
B
D
B
Yeah
criteria
was
that
we
wanted
to
say
that
once
you've
deployed
0.6,
which
is
a
beta
that
going
forward,
we
will
upgrade
you
in
disk
format
and
anything
related
to
block
storage
carries
forward.
So
if
you
were
to
deploy
this
in
on,
you
know,
production,
environment
or
a
staging
environment
that
you
wouldn't
have
to
tear
it
down
and
upgrade
it
manually
or
do
whatever
is
needed.
I
doubt
that
that
investment
carries
for
you
and
that's.
A
B
D
B
D
D
Well,
I
think
anytime,
going
forward.
We're
gonna
have
to
look
at
every
feature
and
say
you
know:
how
can
we
do
this?
So
it's
upgradable
and
every
feature
you
have
to
look
at
the
CR
DS
and
make
sure
that
we
can
change
and
local
storage
will
be
one
of
those
and
we'll
have
to
find
a
way
to
make
sure
there's
a
way
to
upgrade
yeah.
B
B
We
are
promising
them
that
they
don't
have
to
tear
it
down
any
more
houses
0.6
and
that
one.
When
someone
comes
along,
they
have
to
do
some
stuff,
not
automated,
but
it's
we're
carrying
their
data
forward
mm-hmm
and
we're
also
not
changing
the
CRTs
in
a
braking
way.
We
are
we'd,
have
to
document
them
and
upgrade
constructions
so
think
of
it.
From
a
perspective
of
somebody
just
deployed
0.6,
they
started
running
all
these
workloads.
On
top
of
it,
it's
running
fine
point
seven
comes
along.
C
B
A
B
C
B
D
B
The
criteria
we
used
for
beta
was
I
think
exactly
not
somebody
going
from
point
6
to
point.
7
does
not
lose
their
cluster
for
block
storage
great
and
that
we
feel
that
the
CR
DS
4,
which
is
now
our
essentially
our
public
interface.
The
CRT,
is
for
cluster
and
tool
are
ones
that
we
want
to
carry
forward
like
we
would
change
them
from
alpha
to
beta
a
namespace
name
or
whatever
spec
name
or
it
becomes
beta.
B
D
You
know,
I,
don't
know
that
there
are
any
known
major
issues
that
would
block
upgrade
in
those
CR,
DS
I.
Think
there's
just
some
work.
We
need
to
do
with
the
Sierra
T's
like
well,
when
you
update
the
cluster
CRT,
you
add
new
nodes
or
whatever
we
need
to
go.
Property
needs
to
wake
up
and
go
respond
to
that.
I.
Don't
know
why
that
would
impact
upgrade
though,
but
so
and
I
don't
know,
but
I
need
to
change
the
schema
of
the
CR.
These
that's
where
it
would
be
painful.
As
far
as
the
upgrade
okay.
B
Would
be
a
risk
to
calling
it
our
beta
things
like
host
path
and
restart
or
get
a
losing
state
not
because
the
state
is
lost,
but
because
then
it's
not
mangled.
Somehow
I
saw
a
bunch
of
tickets
around
that
Mon
reliability
issues
laughing
moving
too
fast,
losing
IP
issues
that
those
kinds
of
things
there's
still
a
bunch
of
outstanding
tickets,
they're.
B
D
Feel
like
what
the
biggest
piece
missing
is
customer
validation
like
how
many
people
have
actually
tried
this
and
would
trust
their
data
with
it.
Do
we
have
do
we
have
external
people
and
we
have
some
people
playing
with
it
and
trying
it
yeah,
a
major
customer
that
says.
Yes,
we've
tested
this
in
our
environment:
it's
not
just
a
test
environment.
It's
like
real
data
that
that
would
be
awesome
to
be
able
to
declare
with
the
beta
I.
B
A
B
C
Before
we
go
to
0.7,
could
we
talk
about
self
s4c,
lovely
sexy
so
from
the
last
meeting
that
they
had
with
with
you,
Bassam
and
Jared
I?
Think
we
decided
to
use
the
kernel
module
of
self
s
for
now
and
later
in
the
future,
possibly
set
up
for
seven
use
self
use
right.
So
one
thing
about
the
colonel
zareff
s
is
that
support
for
multiple
file
system
is
on
kernel
version.
Four
point:
seven
in
in
after
and
I
I
tested
that
I
tested
this
case
actually
went
to
1604
LTS
the
kernel
version.
C
C
B
C
B
Let
me
just
finish
the
thought
so
when
we,
when
we
were
able
to
do
that,
then
we
can
actually
dynamically
at
runtime
decide
whether
the
capability
is
off
the
kernel
or
whatever
is
needed
on
the
node
support
running
the
best
path.
That
we
know
is
the
fastest
or
more
stable
path,
and
if
not,
we
can
fall
back
gracefully
to
you,
know,
user
mode
or
NB,
dr
BD,
all
that
stuff.
B
B
C
C
So
this
issue
there
is
that
with
the
new
PR
that
travis
submitted
last
week,
when
you
can
create
multiple
file
system,
if
you
were
to
create
more
than
one
file
system
Allah,
so
you
have
two
of
them
and
then
you
try
to
mount
it
from
a
nail
that
does
not.
That
is
zero,
that
for
that's
ordering
4.7,
then
it
will
error
out.
Okay,.
B
C
The
thing
is
that
the
user
will
be
because
when
you
create
two
file
system-
and
then
you
use
the
Nutri
to
attach
it
not
using
the
MDS
the
nsa's,
so
one
of
them
will
get
attached,
possibly
the
first
one,
so
the
user
would
not
know
as
pleasingly
which
one
got
attached
right
out
of
the
two
of
them.
Yes,.
B
C
D
B
B
The
falsest
I
see
are
these:
are
alpha
right,
right,
I,
don't
think
there
is
any
any
issue
I.
The
good
news
is
that
we
have
an
architecture
in
place
to
essentially
be
able
to
dynamically
manage
all
of
us
right.
We
are,
we
will
be
able
to
run
fuse
if
we
decide
to,
and
that
could
be
a
runtime
decision.
All.
B
So
I
went
through
the
issue.
All
the
issues
today
and
I
I
came
up
with
a
bunch
of
themes.
We
should
probably
talk
about
that
I
think,
but
I'm
I
think
will
be
great
if
you
guys
can
all
try
meno.
What
do
you
think
needs
to
happen
in
0.7
if
the
goal
is
to
declare
block
storage
production
really
in
0.7
I?
B
Think
there
is
a
number
of
things.
I
need
to
happen,
all
that
we've
talked
with,
but
I
think
things
around
maaan
reliability.
So
this
is
the
issue
of
the
Mons
failing
over
too
quickly
or
their
based
on
my
timeout
or
the
Mon
map.
Changes
have
not
been
broadcast,
or
even
some
issues
around
mom's
hanging
or
not
restart.
B
D
B
B
I
mean
I,
think
of
it
as
there's
stuff
that
we,
the
operator,
has
a
net
Sydney.
That's
all
the
CR
deeds
in
the
our
state,
and
then
there
are
volumes
underneath
mostly
OS
these
and
moms
right,
and
we
should
be
able
create
a
cluster.
In
fact,
I'd
argue
we
should
be
able
to
kind
of
move
a
cluster
as
a
test
of
this.
A
B
B
No
no
I
mean
I,
certainly
envy
dr
BD
our
higher
priority,
but
I'm.
This
is
the
case
of
I,
run
rook
on
top
of
a
plain
kubernetes
cluster
where
and
it
should
work
regardless
of
kernel
version
or
regardless
of
what's
supported.
It
should
just
work
and
I
think
we're
on
a
path
to
do
that.
But
until
we
get
some
of
the
like,
NBD
are
VD
or
SEF,
Hughes
I
think
we're
still
reliant
on
kernel
versions.
Okay,.
B
D
A
B
So
we
should,
we
should
figure
out
what
the
right
timing
iSight,
given
everything
we've
learned
around.
You
know
the
declarative
management
pattern
of
kubernetes
I.
Think
I
think
we
probably
are
on
a
path
to
say:
there's
CR
DS,
to
manage
the
entire
surface
area
of
Brook,
and
if
you
need
anything
more,
then
you
should
fall
back
to
the
toolbox
and
use
this
SEF
CLI
I.
D
D
C
C
B
A
So
the
one
thing
that
jumps
out
to
me
is
certain
kinds
of
status
like
I.
Think
your
comments
about
the
SEF
monitor
status
is
entirely
correct.
You
should
just
go
to
the
SEF
command
line
or
API
or
something
and
get
it
from
there,
but
do
we
have
a
way
through
the
kubernetes
api
to
get
any
sort
of
health
information
or
metric
information
about
the
storage
cluster
notice.
B
A
B
Or
you
can
imagine
even
status,
monitoring
or
curve
or
all
that
stuff
is
handled
through.
You
know:
post
press
exports
or
not
post
price,
but
Prometheus
things
like
the
Prometheus
exporter
or
others
right
now,
but
in
general
I'm,
just
trying
to
plot
a
path,
a
longer-term
path,
I
yeah,
I,
think
in
general
we
either
I
mean
we
can
go
improve,
see
our
leaves
to
support
better
status
right.
That
is
an
option.
D
C
B
And
it
could
take,
it
could
take.
You
know,
9
releases,
to
get
there
I'm
just
saying
we
shouldn't.
We
should
just
be
clear
about
what
the
roles
are
if
the
API
and
C
liar
and
I
think
it
would
make
sense
to
deprecated,
or
at
least
say
that
we
have
that
this
current
discussion
be
clear
so
that
we
don't
add
more
features
of
the
API
or.
B
B
A
B
A
One
thing
that
I
think
a
lot
about
is
a
cluster
lifecycle,
so
adding
and
removing
devices
adding
or
removing
nodes
I
think
all
those
need
to
be
supported.
It
seems
to
me
that
adding
a
node
is
probably
the
easiest
thing
on
the
list,
but
you
know
how
do
you?
How
does
rook
know
that
a
node
is
going
away?
How
do
we
handle
the
maintenance
life
cycle
around
disk
drives,
etc?.
D
A
B
D
A
B
C
C
B
A
B
B
C
A
B
B
C
C
B
B
D
B
D
B
B
D
B
Then
are
we
saying
that
I
think
we
talked
about
performance
testing
as
being
a
beta
thing,
we're
not
going
to
focus
on
performance
until
we
till
we
declare
some
things
beta,
I'm
assuming
then
performance,
testing
for
block
storage
becomes
it's
somewhere
here
right
or
else.
How
do
you
get
to
production.
C
B
B
B
B
B
B
B
C
B
D
Maybe
a
clarification
on
the
tickets
were
looking
out
here.
There
are
things
that
I
think
block
o
dot.
Six.
Are
there
things
that
they're
little
things
and
we
will
fix
them
in
the
next
two
weeks
type
of
thing?
How
can
we
differentiate
between
these
because
I
think
this
one
is
one
of
those
were
it's
not
required
because
running
cuvee
diem,
where
are
like
is
working.
A
A
C
C
B
C
C
A
B
Okay,
all
right
we're
almost
out
of
time,
I,
think
I,
think,
let's,
let's
go
go
through
your
tickets
and
update
the
ones,
especially
you
Stevie,
because
you've
got
the
biggest
change
here
and
then
in
the
next.
So
the
next
meeting
is
a
two
weeks
from
now.
Let's
make
a
call
on
whether
point
six
is
ready
and
what
needs
to
be
bumped
out
or
not
sounds
good.
All
right.
Thanks.