►
From YouTube: Keynote: Town Hall - Panel
Description
Keynote: Town Hall - Panel
This will be a town hall panel with the Ceph Component leads. Please submit questions ahead of time to the etherpad, or ask them during the session. https://pad.ceph.com/p/cephalocon-2019-town-hall.
A
Well,
I'll
be
moderating
that
you
through
pad
from
down
here,
but
please
welcome
the
component,
leads
and
stern
repeaters.
So.
B
Good
morning,
this
is
your
chance
to
ask
them
all
the
questions.
You've
been
dying
to
ask
them
for
the
last
year,
but
this
isn't
your
only
chance
so
ask
questions
here.
You
can
enter
them
in
the
ether
pad,
but
also
catch
them
during
the
conference.
So
make
sure
you
get
your
questions
answered,
I
mean
here's
your
chance,
I'm
gonna,
start
out
by
having
everybody
introduce
themselves
and
I
was
trying
to
think
of
an
interesting
question
and
there's
like
the
one.
That's
the
perennial
debate.
B
C
D
D
E
F
A
G
C
B
Obviously,
people
share
your
opinion
I'm,
so
on
the
etherpad
you
can,
you
can
enter
in
questions
I'm
going
to
try
to
ask
I
am
not
sure
I'll
be
able
to
get
to
all
of
them,
but
we'll
ask
some
of
them
and
we'll
try
to
answer
others
later.
If
you
have
questions,
keep
adding
them.
If
I
ask
your
question
wrong,
feel
free
to
jump
up
and
wave
your
hands
and
say:
that's
not
it,
but
there
are
a
couple
of
questions
about
how
to
get
started.
B
So
if
somebody
came
to
the
conference
they've
been
following
stuff
for
a
while
they've
been
using
it,
they
have
questions
about
how
to
use
it
better
and
now
they
want
to
contribute.
How
would
they
start
contributing?
Maybe
a
couple
of
you
could
give
a
give
a
suggestion
for
how
to
get
started.
How
do
you
find
something?
That's
manageable
for
a
new
person
to
start
with
I.
A
Personally,
I
think
one
of
the
the
nicest
and
easiest
ways
to
get
started
and
get
acclimated
to
the
code
base
would
be
to
help
with
the
the
back
porting
effort
when
we
fix
bugs
in
the
in
the
current
release
of
software.
That
same
block
might
affect
previous
releases.
The
software
and
patches
don't
necessarily
apply
cleanly.
G
G
Other
than
that,
we
have
a
couple
of
other
things
that
we're
doing
is
we
have
this
effort
called
doc,
you
better,
which
is
a
way
in
which
you
can
contribute
to
the
docs,
which
is
a
simple
way
that
you
can
get
started
with,
contributing
to
and
even
create
pull
requests.
So
we
have
now
and
edit
on
github
link
on
the
docs
page.
G
If
you
find
a
bug,
you
can
go,
just
fix
it
even
without
going
through
the
whole
process
of
you
know,
creating
and
cloning
a
repo
and
doing
all
that
stuff
and
the
other
thing
I
think
we've
almost
done
it
and
we
started
doing
it,
at
least
as
in
our
red
mind
tracker.
We
have
also
started
including
a
tag
of
for
bugs
with
low
medium
and
high
difficulty
levels.
So
if
you
find
anything
which
is
low
medium,
just
go
pick
it
up.
Let's
say
I.
H
Think
one
thing
I
want
to
add
is
that
part
of
the
reason
why
the
learning
curve
for
stuff
is
steep
is
because
it's
at
some
well
at
some
level
it's
just
complicated
and
it's
written
in
C++.
Not
everybody
is
a
C++
developer,
but
since
I
can
he
was
crackin.
Actually
we
have
the
new
stuff
manager
that
has
the
whole
Python
runtime
and
a
lot
of
the
newer
and
like
visible
features,
are
the
integrations
with
Seth
and
that
automated
management
functions
integrations
with
Prometheus.
H
E
And
I
also
want
to
add
that
we
have
even
more
space,
because
we
we
have
space
for
web
developers
to
contribute
to
safe
word.
So
it's
currently,
it's
angular
2,
so
yeah
complex
itself
is
a
very
big
project
and
has
so
many
components
with
you
that
are
developing
different
languages.
So
there's
space
for
almost
any
kind
of
programmer
and.
B
H
Yeah
I
think
I
mean
doc
there.
There
are
a
couple
sort
of
problem
areas
or
challenges,
I
would
say
with
documentation.
The
first
is
that
the
information
architecture,
the
structure
the
documentation,
was
conceived
and
put
together
like
five
six
seven
years
ago,
and
it
hasn't
really
been
updated
since
then,
so
at
some
level
somebody
needs
to
like
sit
down
and
say
like
we
should
just
organize
this
differently
and
I
think
there
are
a
lot
of
models
to
look
at
most
other
successful,
open
source
projects.
H
A
lot
of
the
documentation-
that's
severe,
is
just
it's
just
out
of
date,
so
just
reading
the
documentation
and
anybody
who
is
a
current
operator
or
stuff,
it
has
sort
of
a
sense
of
like
what
the
current
state
of
the
state
of
the
art
is
just
reading
through
the
documentation.
You'll
very
quickly
see
that
the
click-
oh,
this
isn't
actually
quite
what
it
should
be
now
and
those
edits
are
super
easy
to
do
because
of
the
new
new
github
link
on
every
page.
B
Awesome
so
we
won't
update
the
docs
and
all
of
you
can
help
alright,
so
the
audience
can
help
me
out
if
you're
entering
in
questions
that
there's
no
way
we're
gonna
answer
all
these
questions
in
this
half-hour
session
I'm.
So
the
people
that
are
adding
plus
ones
to
questions,
that's
awesome,
so
add,
plus
ones
that
the
questions
do
you
want
to
see
answered?
Well,
we
have
deep.
This
is
a
tongue
twister.
Well,
we
have
deep,
deep
duplication
in
the
set.
H
H
Certainly,
yes,
at
some
point:
I'm,
not
gonna,
not
gonna,
make
any
promises
as
far
as
when
it
when
it's
going
to
happen,
but
there
is
ongoing
work
with
deduplication
and
there's
a
group
at
SK.
That's
been
working
diligently
at
this
I
think
that
the
and
there's
a
there's
a
general
view
of
how
it's
going
to
be
implemented
so
that
it's
actually
distributed
you
do
so,
if
you
have,
if
you're
right
and
in
a
scalable
way,
if
you
write
the
same
data
twice,
it's
always
gonna.
Do
you
up
to
one
thing?
H
If
you
try
to
read
the
data
later,
and
so
what
we're
focusing
on
right
now
is
creating
a
tool
that
will
let
you
scan
an
existing
data
set
with
different
chunking
parameters,
to
try
to
figure
out
what
the
actual
ratios
are,
that
we're
gonna
get
in
practice,
and
that
and
once
we
have
that
available,
and
we
get
real
users
to
run
that
on
real
data
sets.
We'll
have
a
better
idea
of
what
the
what
the
benefits
and
same
is
going
to
be
and
that'll
help
us
prioritize
and
focus
our
efforts.
B
A
So,
from
from
an
orbit
of
view,
just
the
general
rule
of
thumb
is
is
care,
BD
is
probably
gonna,
be
faster.
Can
ish
percent
faster
depending
on
your
workloads,
but
it's
always
gonna
lag
behind
in
terms
of
what
features
it
can
support
just
due
to
the
complexity
of
getting
changes
into
the
kernel
and
just
the
lead
time
getting
those
kernel
updates
to
various
operating
systems.
We
nowadays
on
the
on
the
docs
page,
we
didn't
used
to
have
a
good
mapping
about
what
are
we
do.
A
Features
specifically
were
supported
by
what
versions
of
the
kernel
we're
trying
to
be
better
at
that.
We
do
now
have
some
documentation
to
say
here's
the
upstream
kernel
version
where
these
features
are
available
in
the
KB
view,
driver
that
doesn't
apply
to
things
like,
but
at
a
PI's
Linux
where
things
are
back
ported,
but
it's
not
you
know,
they're,
not.
Those
are
features
are
not
in
the
3x
kernel
that
Red
Hat
is
based
off
of
in
terms
of
when
to
use
what's
right
now,
a
lot
of
those
choices
are
probably
made
by
your
orchestration
tool.
A
You
know
if
you're,
using
libvirt
sand
or
directly
and
your
your
Tynan
VM
images.
It's
it's
gonna
use
live
RVD
community
use
live
RVD
if
you're
using
kubernetes
in
the
CSI,
it's
going
to
be
using
care
BD,
because
cabanne
needs
a
block
device.
We're
talking
about
the
NBD.
That's
the
network
block
device,
there's
the
RVD
driver
for
that
that
uses
live
RVD.
C
Yeah
force
FS
most
of
what
Jason
said
about
our
BT
also
applies
to
Southwest.
Generally,
we
try
to
keep
feature
parity
between
the
kernel
and
SEF
views.
Really
when
we
say
SF
used
were
talking
about
lips
ffs,
which
is
also
being
used
by,
for
example,
Ganesha
when
you're
exporting
stem
fest
with
NFS.
C
More
recently,
we're
actually
looking
at
adding
performance
improvements
and
focusing
on
the
kernel
sage
was
talking
earlier
about
how
we're
looking
at
doing
a
synchronous
on
Lincoln
create
and
that
work
is
being
focused
in
the
kernel
for
mostly
technical
reasons,
but
also.
We
think
that
the
improvements
are
gonna
be
more
impactful
in
the
kernel
rather
than
in
the
user
space
client
library.
Right
now,.
B
A
Yeah
we
at
least
from
our
point
of
view,
I
I,
still
talk
to
maintain,
errs
and
I,
wash
the
PRS
or
their
equivalent
of
tractor
tickets
that
come
in
against
the
cinder
drivers.
So
it's
it's
still
actively
being
developed,
it's
still
actively
being
maintained
and
so
and
I
think
we
have
a
good
back
and
forth.
It's
just
a
little
bit
different
environment.
Now,
because
it's
more
of
a
mature
product-
and
you
know
we're
not
the
ones
driving
the
the
innovation
and
integration
they're
they
now
have.
You
know
a
good
team
and
they're.
A
You
know
trying
to
integrate
features.
You
know
directly
like
when
cinder
atted,
the
the
multi-site
replication
feature.
They,
you
know,
help
pull
in
that
feature
and
tie
that
in
to
RVD
nearing
a
multi
I
attached
same
thing
that
was
driven
from
the
cinder
point-of-view
now
for
most
pushing
the
feature
to
them.
H
It
was
a
long
time
ago
now,
probably
most
ten
years
ago,
but
I
spent
like
a
week
trying
to
construct
a
workload
and
a
benchmark
that
would
actually
show
a
performance
improvement
by
reading
from
multiple
caches
and
I
I
couldn't
do
it,
and
it
really
means
that
your
your
bottleneck
has
to
be
like
get
to
be
saturating.
Individual
mix,
I'm,
single
replicas
and
those
types
of
work.
G
B
G
We
do
agree
that
our
problems
are
with
rocks,
TV,
leveldb
I
think
we
don't
use
anymore
so
that
pain
is
gone,
but
I
think
we
are
trying
to
find
ways
around
the
existing
problems.
But
we
are
not
really
I
mean
like
we
have
some
contributions
that
have
been
made
to
upstream
rocks
TV
but
they've
been
I,
mean
the
encouragement
of
acceptance
of
pull.
Requests
have
been,
you
know,
a
challenge
for
us,
I
guess
so.
I
guess
we
internally
are
trying
to
fix
issues
by
doing
things
differently,
rather
than
fixing
the
codebase
of
rocks
to
be.
H
B
H
Would
add
just
that
there
are
a
lot
of
very
successful,
very
large-scale
Swift
opponents
out
there.
There
are
several
public
clouds
that
use
rgw
to
back
the
rest
free
service,
and
there
are
also
several
very
large
foot.
The
clouds
that
use
Swift
to
back
their
object,
storage
and
I-
don't
see
them
rushing
to
change
technologies,
it's
working
for
them,
so
more
power
to.
B
All
right,
so
there
was
a
question
I'm
one
of
the
early
people
asked
how
you
all
work
together.
So
when
you're,
when
you're
thinking
about
questions
like
this
difficult
questions
for
the
project
moving
forward,
what's
gonna,
you
know
what
you're
gonna
put
your
resources
on.
How
do
you
all
work
together,
I
gotta,
on
a
given
day?
How
do
you
communicate
with
each
other?
D
H
Think
the
only
two
of
us
that
are
in
the
same
office
are
Patrick
and
Neha
and
I
view
that,
as
almost
a
coincidence
for
the
most
part,
that
team
is
very
distributed.
So
we
use
a
lot
of
I
spent
half
my
morning
on
blue
jeans
in
the
various
stand-ups
and
synchronization
clothes
just
to
like
check
in
on
what
pull
requests
are
in
flight.
What
we're
working
on
what
our
focus
areas
discuss
whatever
random
issue
has
come
up.
So
a
lot
of
those
things
so.
B
G
So
sometimes
we
have
short
meetings
on
blue
jeans
and
maybe
are
not
convinced
about
things.
We
follow
it
up
on
IRC,
and
this
probably,
like
you
know,
conversations
going
on
for
half
an
hour
and
out
of
them
are
like
you
know,
just
between
like
two
people,
just
trying
to
you
know
sort
it
out
and
trying
to
come
to
a
conclusion.
So
I
think
IRC
helps
a
lot
that
way
and
it's
it's.
C
Yeah
I
just
wanted,
you
know
because
we're
a
global
team
and
cross
company.
You
know
we
really
do
have
to
embrace
the
open
source
methodologies,
and
that
means
using
like
the
mailing
lists
and
and
get
up
in
public
code
review.
There's
no
real
other
way
to
do
it.
We
don't
have
you
know
secret
back
channels
where
we
do
development
and
argue
generally
during
meetings
that
we
do
have
over
the
phone
or
on
blue
jeans.
We
don't
usually
disagree
too
much
if
we're
gonna,
you
know,
discuss
some
code
at
length.
C
C
B
G
B
C
C
You
know
coming
this
end
up
a
lot
of
projects,
so
we
do
work
on
sometimes
there
long
shots
and
it's
difficult
to
predict
when
they'll
actually
finish,
but
otherwise
one
of
the
tools
you
can
look
at
to
kind
of
get
an
idea
of
what
our
priorities
are
is
to
look
at
the
Ceph
tracker
and
look
at
the
next
release,
the
target
release
field
and
the
tracker
for
all
that
and
do
a
search
that
I'll
kind
of
give
you
a
rough
idea
of
what
our
priorities
are.
Some
some
tickets
could
probably
have
their
priorities
refreshed.
B
B
E
So
for
safe
dashboard,
of
course,
we
are
always
working
on
improving
usability,
but
our
main
focus
for
the
future
is
the
integration
with
the
artist
writer
and
allow
for
the
users
to
do
the
day.
2
operations
like
replace
discs
or
adding
OSDs
or
any
other
service,
and
we
always
have
this
work
of
keeping
up
with
new
features
that
appear
in
another
component
right.
If
they
are
well
suited
for
being
managed
by
a
GUI.
We
need
to
make
sure
that
we
we
plan
that
and
we
implement
those
so.
B
E
B
A
So
to
follow,
along
with
what
we
have
planned,
that
we
actually
have
a
Trello
board,
it's
a
set
back
log
and
what
we
actually
also
have
labels
on
everything
is
categorized.
You
know
here's
things
for
corridos,
here's
things
for
RVD,
here's!
You
know
things
for
rgw
on
the
our
beady
thick
board.
We
have
everything
labeled
saying
like
for
this
upcoming
octopus
release.
You
know
everything
that
we
want
to
try
to
tackle
for
this
release.
We
have
it
tagged
things
that
have
already
been
merged.
We've.
A
Hardly
you
know:
they've
been
merged,
moved
to
the
merged
in
octopus
board',
but
for
a
high
level
of
what
we're
planning
for
this
next
release.
On
the
kernel
side,
cavity
we're
trying
to
just
keep
improving
feature,
parity
I
think
the
goal
would
be
with
the
5.3
kernel
will
pretty
much
have
everything
are
all
our
BT
features,
with
the
exception
of
journaling
and
the
new
live
migration
feature
that
got
added
in
Nautilus
on
the
lines
of
the
live
migration?
We
want
to
extend
that
some
more
right.
Now.
A
It's
it's
live
migration
of
images
within
the
same
cluster.
We
want
to
extend
that
so
that
it's
it's.
We
want
to
provide
migration
capabilities
between
different
clusters,
different
clouds
of
a
of
workloads
for
block
and
just
in
general,
also
performance
and
the
lib
RVD
side.
I've
mentioned
before
it's
it's.
It's
got
performance
overhead.
That
makes
it
a
little
bit
slower
than
caridy.
G
Me
so
I
think
I
have
the
same
thing
to
say
as
Jason
most
of
the
broad
level
tracking
in
Raiders
happens
in
Trello
and
the
way
we
decide
what
we
are
going
to
be
doing
in
a
particular
release
is
probably
discussed
in
a
forum
like
a
CDM
which
we
do
try
to
like,
say.
Ok,
this
is
what
we
are
at
least
going
to
try
for
octopus
and
what
is
realistic
and
what
is
not.
And
then
there
are
obviously
smaller
meetings
on
blue
jeans
and
sometimes
stand-ups
just
15
minutes
Randolph.
G
They
just
become
45
minutes
standards.
Just
discussing
you
know
deeper
things,
but
I
would
also
like
to
add
that
for
crimson
we
are
using
a
separate
github
project
and
the
tracking
there
is
more
granular
about
what
what
are
the
broad
level
things
that
we
want
to
do
and
what's
getting
merged
every
day,
and
it's
pretty
pretty
neat.
Actually,
you
know
every
pod
requests
that
gets
merged.
You
can
see
it
that
this
feature
was
something
that
we
was
are
working
on.
It's
now
much
the
tracking
becomes
really
easy
there
and
other
than
that.
G
H
Yeah
I'll,
just
repeat
the
the
Trello
thing,
so
I
try
to
delegate
as
much
as
possible
to
these
folks.
I-I-I
can't
keep
track
of
everything
and
it's
the
project
has
scaled
much
beyond
what
I
ever
could
have
wanted
would
have
wanted
to
have
my
hands
in,
and
so
the
challenge
for
me
is
just
keeping
in
synch
with
the
rest
of
the
the
leads
checking
with
him
periodically
and
finding
out
what
what
their
plans
are
and
what
their
focus
is
and
the
way
that
I
am
able
to
do
that
is
through
that
Trello
board.
H
But
we're
in
the
process
of
revising
we'll
have
links
to
accept,
for
example,
the
Trello
board.
So
anybody
can
go.
Look
it's
a
public
Trello
board,
so
you
have
to
have
be
a
member
to
edit
it.
But
you
can
go.
Look
at
what
what
all
the
stuff
is
and
as
things
get
merged,
they
get
plopped
into
the
novelist's
or
the
octopus
category.
So
you
can
see
progress
as
we
move.
B
So
thanks
for
being
brave
and
being
up
here,
if
you
didn't
get
your
question
answered
now,
you
know
what
they
look
like,
where
to
find
them
they're
here
for
the
rest
of
the
day,
I
think
at
least
one
of
you,
a
couple
of
you
still
have
talks
today,
who
has
a
talk
today
or
cuddle?
Ok,
so
if
you
want
to
hear
more,
there's
another
talk,
so
please
find
them.
Please
talk
to
them.
Please
thank
them
for
their
work
and
thank
you
for
all
your
questions.