►
From YouTube: 2016-SEP-07 :: Ceph Developer Monthly
Description
Monthly developer meeting for the coordination of Ceph project development.
http://wiki.ceph.com/Planning
A
More
can
please
welcome
back
to
the
the
monthly
staff
developer
meeting
here
looks
like
we
got
a
pretty
good
crew,
but
will
be
probably
have
a
short
one
tonight,
and
we
have
a
whole
lot
of
a
lot
of
blueprints
on
the
docket
today.
So
sage,
where
you
want
to
start
with
this,
when
you
wanna
start
with
your
messenger
to
plans,
let's.
D
Okay,
all
right,
so
this
this
is
the
blueprint
I
had
this
I
would
like
to
discuss
today.
I
would
like
to
discuss
snapshots
and
consistency
groups.
So
far,
we
have
to
pull
requests
that
have
been
merged
in
this
pull
request,
included
operations
for
creating
deleting
and
wasting
snapshots
operations
for,
adding,
removing
and
listing
images
and
not
snapshots
consistency
groups,
then
adding
removing
wasting
images
in
things,
consistency,
groups
and
now
I
would
like
to
discuss
design
for
snapshots
of
consistency
groups.
D
D
Snap
images
snapshots
so
these
are
like
that
does
go
to
f
with
you,
snapshot
underscore
and
snapshot
ID
in
every
image
and
I
suggest
to
add
this
field
for
this
image
of
snapshot,
which
is
called
snapshot.
Oh,
but
actually
right
so
in
this
snapshot
type
will
have
taste,
not
five
snapshot.
Preference
will
have
this
type
boost
variant.
D
One
of
two
self
sending
snapshot
of
group
members,
natural
soft
enemy
snapshot,
adjusting
an
empty
constructor
with
no
belly,
just
specifying
the
yellow
so
specifying
that
this
is
a
lake,
a
snapshot
that
was
created
individually
and
group
number
snapshot.
It
means
that
this
snapshot
was
created
as
part
of
group
snapshotting
process,
and
then
this
group
member
snapshot
will
have
fields
that
allow
us
to
identify
what
group
snapshot.
D
It
belongs
to
group
pool
ID
group,
ID,
nutshot
ID,
so
yes,
and
as
a
result,
we
have
to
modify
decode
and
encode,
but
only
so
I
just
borrowed
this
code
code
from
other
decoding
and
encoding
first
for
similar
data
structures.
D
So
this
is
how
we
want
to
modify
image,
snapshots
image
snapshots,
to
allow
referencing
group
snapshots,
and
in
order
to
create
these
snapshots,
we
need
to
be
able
to
specify
who
created
them.
So,
as
I
see
this,
we
will
add
one
more
parameter
to
snap
free
function
in
image
class
and
this
type
will
have
this
snapshot
type
of
snapshots,
reference
and
type
which
yep,
basically,
if
its
users
snapshot
them
by
default
movie
is
a
snapshot.
Otherwise
it
will
be
good
snapshot.
So.
B
E
Been
changing
them
to
users
now
she'll,
so
that
would
be
like
the
old
enough
field
legacy
style.
I
was
like
it
was
just
something
that's
created
to
the
APR.
Okay,
like
no
external
links,
okay,
not
only
not
in
a
namespace
convention.
It's
you
know,
user's
name
space
as
opposed
to
oh
and
as
something
that
I
feel.
This
is
something
that's
gonna
have
to
be
used
also
for
group
snapshots
and
where
you
would
also
want
to
recycle
this
capability.
For
the
sync
point
snapshots
that
are
created
by
our
BD
mirroring.
E
This
is
something
that
kind
of
fell
off
the
table
as
it
was
nice
to
have,
but
never
we
never
got
it
implemented
for
our
BD
mirroring
okay,
so
it
causes
some
problems
with
already
nearing
to
or
if
you
try
to
delete
an
image
where
a
sink
has
just
started.
You
hadn't
put
like
you're
racing
with
a
sink
point
snapshot
that
still
exists,
but
it
was
something
that
wasn't
created
by
the
user.
We
should
be
able
to
handle
it.
E
B
So
oh
I.
D
B
D
B
It
it's.
D
So
this
is
how
I
suggest
to
create
special
snap
shots
with
by
just
adding
the
this
parameter.
Yeah,
I'm
not
sure
if
it's
the
right
thing,
but
so
far
it
seems
to
me
they.
This
is
a
idea.
If
it's
wrong,
open,
yeah
I
go
up
to
here,
die
opinions.
So
what
does
it
mean
right
so
big?
Because
because
of
because
of
because
of
this
will
want
to
expose
oh
I,
I
suggest
to
expose
some
of
methods
that
are
available
in
image
context,
but
not
available
in
image.
D
D
D
As
I
see
the
implementation
I'm
going
to
grade
this
function,
create
image
snapshot
well,
because
it's
great
consistency,
group
snapshot
as
a
function
in
with
barb
ed,
/
internal,
see
file
so
I
when
I
start
creating
snapshot.
I
have
all
images
that
are
in
this
consistency
group,
so
I
open
those
images,
yeah.
B
D
E
Would
know
because
because
so
you
you
have
a
new
liver,
an
internal
in
internal,
so
you
have
an
API
method,
like
consistency
group,
snap,
snap
creator,
something
like
that
groups?
Yes,
but
you
know
which
cult,
which
calls
an
internal
method
that
it
has
to
iterate
through
all
the
images
that
are
associated
with
that
consistency,
group
and
as
part
of
iterating,
through
that
it
has
to
open
up
all
those
images
internally.
But
it's
not
opening
those
images
via
the
lib
rbd
external
public
API,
its
opening
them.
E
Don't
have
to
worry
about,
you
know
ABI
or
API
breakages,
because
the
color
won't
compile.
You
know
yeah.
D
E
G
E
And
then
once
it
open
all
once
it
opens
all
of
them
that
it's
going
to
try
to
grab
these
scoops
a
block
for
all
of
them
and
then
once
it
grabs
the
occlusive
lock
for
all
of
them
again.
This
is
all
just
directly
invoking
methods
chained
off
of
lib
rbd
image
context
I'm.
So
like
image
context,
state
open
image,
context,
exclusive,
lock,
trilok
I
should
much
luck.
So
once
you
have
all
the
images
opened
all
the
locks
acquired,
then
you
can
basically
just
go
straight
through
and
go
image
image
context.
Operation
snap
creates
passing
in.
E
You
know
this
this
this
snapshot
namespace
or
reference
to
your
your
group,
and
then
you
kind
of
collapse.
All
these
things
together,
because
you
can
spawn
this
off
in
parallel,
I
mean
once
you
get
all
you
can
open
all
the
images
in
parallel.
You
can
grab
all
these
suits
of
blocks
in
parallel
and
then
you
can
snapshot
create
in
parallel
up
to
up
to
the
rbd
num
concurrent
ops
and
there's
helper
functions
out
there
to
help
you
throttle
I'll.
E
You
know
he's
kicking
off
state
machines
up
to
up
to
that
limit,
and
so
then,
once
once
all
those
then
finish
up
that
now
you've
you've
you've
gotten
all
your
images
open.
They
have
these
susa
block
and
I've
created
all
your
snapshots.
So
you're
done,
then
you
close
all
your
images.
If
you
have
a
failure,
any
point
in
time
I
mean
if
you
couldn't
open
any.
E
If
you
couldn't
open
one
of
the
images
abort,
if
you
couldn't
grab
the
exclusive
lock
fall
in
the
images
abort,
if
a
snapshot
failed
to
be
created
on
one,
do
you
abort
or
to
use
return,
air
and
force
the
user
to
say
group
snap
remove
to
get
to
a
clean
state?
E
The
problem
of
the
boarding
is,
you
could
still
crash
in
the
middle
or
something
like
that.
So
you
could
no
matter
what
you
you've
got
to
be
able
to
handle
the
case
of
snap
group
snap
remove
and
it
has
like
some
images
that
have
the
snapshot
and
some
images
the
down.
It's
just
got
to
be
able
to
say
yeah
I've
cleaned
up
successfully.
D
E
Good
I
mean
that
the
concept
is
I,
think
you
know
the
at
the
high
level
on
the
CLS
side,
the
class,
the
cost
method
side.
We
want
to
have
a
new
data
structure,
variant
I'd,
like
all
that
stuff
be
able
to
hold
these
things.
The
one
comment
I
made
is
just
because
it's
it
probably
would
be
good
to
enforce
a
new
feature
bits
on
a
consistency
groups
on
a
given
image,
because
you
do
you
don't
necessarily
want
an
older
clients
that
doesn't
understand
all
these
new
rules
about
how
to
handle
consistency.
E
Groups
to
be
able
to
you
can
open,
make
an
open
it
read-only,
but
there's
a
special
flags.
They
say
like
hey.
If
you
want
to
open
it,
as
you
know,
a
read/write
image
you
need
to
support,
you
know
these
future
boots.
So
I,
that's
one
of
the
emails
I
sent
a
while
ago
said.
I
think
we
we
do
want
to
have
a
consistency
group
feature
bit
and
it's
just
a
dynamic
feature,
but
it
can
be
turned
on
and
off
dynamically.
E
As
you
add
an
image
to
a
consistency
group
or
remove
an
image
from
a
consistency,
group,
I
think
that's
important,
I
mean
I,
think
you're
on
the
right
track.
My
point
of
view.
D
E
A
E
It
protected
be
a
large
method,
so
just
I
sort
of
I've
been
taking
things
out
of
internal
and
moving
them
into
like
here
in
the
segregated
locations
of
where
things
go
like
everything,
here's
everything
I
deals
with
I.
Oh
here's,
everything
that
deals
with
operations,
so
this
would
be
something
that
does
with
consistency.
Groups
it'll
be
nice,
and
this
up
something
has
I
haven't
mentioned
before,
but
I
think
could
be
good
to
keep
pushing
that
direction
of.
E
F
D
D
E
So
well,
you
could
you
could
handle
it
in
one
or
two
ways
so
that
when
you
are,
when
you
basically
owned
the
Susa
block,
you
can
optionally
say:
I'll
accept
requests
from
other
other
clients
or
a
bit
or
I'll
reject
request
some
other
clients.
I
would
request
you.
E
E
You
grab
the
exclusive
block
and
I
your
weight
on
image,
a
and
now
you're
waiting
to
get
things
with
lock
an
image
be,
and
somebody
sends
your
request
saying:
hey
can
I
have
an
exclusive
lock
back
and
you
and
then
in
the
background,
says
yeah
sure
have
that
exclusive
lock
that
because
you
requested
it,
if
it's
probably
just
easier
to
say,
I'm,
not
accepting
requests
right
now
and
it'll
automatically
in
the
background
handle
like
telling
those
clients-
hey,
I'm,
not
I'm,
not
accepting
requests
right
now
and
and
they'll
deal
with
and
they'll
keep
sending
or
they'll
keep
trying
again
and
again
and
again
until
you
are
sending
requests,
but
in
this,
in
this
case,
in
a
perfectly
functioning
system.
E
E
E
On
the
other
side,
you
have
u
fq
processes
or
whatever
that
now
I
have.
I
oh
blocked.
I
mean
you
don't
want
this
to
be
along.
We
shouldn't
design
this
to
be
0
or
worry
about
this
being
like
a
long
enough
living
state
that
we
need
to
worry
about
handling.
You
know,
requests
from
other
people
because
once
you
start
handling
requests
for
other
people,
if
you
do
it
fast
enough,
then
you've
got
to
basically
abort
those
requests
to
say.
I
finished
that
I'm,
giving
up
the
lock
so
hey
go.
E
Oh
I'll
put
some
code
examples
in
your
in
your
pad
about
or
point
you
to
where
we
do
that
we
block
for
questions
like
that.
Now.
D
That
would
be
perfect.
Okay,
another
thing
I'm
curious
about
so
when
we
create
this
snapshot,
but
my
like
first
idea
was
to
do
when
we
create
this
consistency,
groups
snapshot
and
then
we
create
individual
snapshots.
We
have
to
give
those
individual
snapshots
names
and
I
was
thinking
that
this
name
should
be
the
same
as
consistent
group
snapshot,
but
then
I
realized.
Maybe
it
makes
sense
to
do
them
completely
random
so
that
they
do
not
clash
with
any
existing
names.
D
E
If
anything
I
think
they
don't
have
a
name
and
bear
with
me
on
this
because
you
cannot
live
our
buddy
can
look
up
the
name.
If
it's
got
that
snap,
it's
got
that
consistency
group
linkage.
If
I
do
a
list
snap,
it
could
look
up
the
name
to
give
Phil,
because
it's
an
ID
right,
because
I
can
rename
a
snapshot
so
instead
of
having
basically
just
a
random
ID
that
says
this
is
snapshot.
Xyz
and
you
don't
know
what
two
links
back
to
I.
E
G
E
Lyst
two
or
something
like
that,
where
we
can
kind
of
delineate
or
you
know,
provide
those
extra,
you
know
pieces
of
metadata
about
saying,
like
hey:
here's,
a
snapshot,
here's
his
name's
XYZ
and
it's
namespaces.
It
was
a
user
created
snapshot
or
here's
one
that
was
created
by
its
name
as
XYZ
and
it
was
created
by
our
body
mirroring
or
here's
one
named
XYZ
and
it
was
created
by
consistency
groups.
So
when
you
do
a
snap
list
on
the
rbd
CLI,
you
can
have
like
an
extra
column,
but
bold,
maybe
even
by
default.
E
Probably
potentially
is
it
doesn't
even
show
you
things
that
aren't
created
by
the
by
the
user,
because
why
do
you
care
because
they'll
automatically
get
cleaned
up?
It's
just
going
to
be
confusing.
If
you
see
like
these
system
internal
snapshots,
but
maybe
you
could
do
like
a
dash
dash,
verbose
or
something
like
that
on
snap
list
that
you
can
see
all
the
internal
of
you
know.
Here's,
although
here's
the
group
snapshots
and
here's
the
harbor
demian
snapshots
and
exercise
they
just
a
thought.
E
Yeah,
yeah
or
or
least
provide
enough
metadata
on
the
CLI
to
say
like
this
was
user
created.
This
is
created
by
a
consistency
group.
This
was
created
by
you,
know,
I,
think
that'd
be
a
nice
feature
so
case
one
doesn't
say
like
especially
if
it's
like
the
consistency
group
snapshot
name
is
foo
and
you
see
a
snapshot
name
foo
you're
like
whoa,
where
did
that
come
from
and
I
can't
delete
it
and
because
I
have
to
delete
it
to
the
consistency
group
I
can't
delete
it
off
the
image.
It's
you
know
directly
right.
D
J
E
D
Okay,
ok
and
I'm
not
a
push.
D
Alright,
so
here's
the
thing
so,
for
example,
we
do
we're
in
the
state
group
safe
committing.
It
means
that
all
snapshots
are
done.
They
recorded
in
the
only
part
that
is
missing
to
rename
this
candidate
to
normal
snapshot.
The
my
suggestion
is,
to
just
say,
hey,
try
again
and
completely
separation
rather
than
rolling
business
back,
oh
right,
because
if
all
snapshots
are
done-
and
the
only
part
is
that
is
missing
is
to
rename
this
attribute
yeah.
E
I,
wouldn't
worry
about
any
of
these
things
is
kind
of
like
the
rest
of
the
liver,
ed
operations.
I
wouldn't
worry
about
rolling
them
back.
I
would
worry
about.
Candy
is
there.
It
is
the
kind
of
like
the
opposite
operation.
Able
to
handle
a
partial
state
like
I
can
partially
create
an
image.
I
mean
there's
so
many
steps
that
involved
in
creating
an
image,
but
any
number
one
of
them
can
fail.
I
don't
go
and
create
an
image
again.
I,
don't
worry
about
trying
to
when
I
open
the
image.
E
E
They
would
remove
it
and
start
again,
because
even
if
you
crashed
while
you,
you
know,
you
got
one
out
of
n,
you
know
snapshot,
you
know,
sudden
inch
snapshots
created
you're
going
to
be
in
the
same
state
of
you
know:
I'm
just
I
got
them
all
created
and
I
just
haven't
flipped
the
bit
that
says
you
know
I
verified
that
everything
is
good.
So
you're
kind
of
in
a
pending
state
too.
Yes,
everything
is:
is
this
good
state.
E
D
E
What
especially
for
first
pass
I
wouldn't
worry
about
all
these
cleanup
states
who
I
think
there's
gonna,
be
so
many
edge
conditions
on
these
cleanup
states
that
it's
probably
just
easier
to
say,
I've
started
to
create
it.
You
know
depending
it
or
it's
complete,
and
if
it's
pending
it
could
have
been
anywhere
along
with
you
know,
along
the
lines
of
creating
it
and
beside.
E
D
D
E
D
H
D
B
All
right,
let's
afraid
us
still
out
here,
so
let's
do
the
the
kerberos
ation
discussion
with
a
band
right
here:
Dan
yeah.
B
Yep,
so
I,
just
I
was
just
trying
to
find
an
etherpad
to
take
notes
and
stumbled
across
an
old
etherpad
from
I
think
of
session
a
long
time
ago.
So
I
puts
on
the
chat
and
I
think.
Actually
this
is
the
most
specific
detail.
That's
been
written
down
as
far
as
what
what
sort
of
that
first
set
of
steps.
Are
you
want
to
take
a
look?
G
I
L
L
L
Well,
I
know
other
members
of
the
crab
members
of
the
other
of
the
Kerberos
kriti
would
say,
but
it
seemed,
but
I
have
to
say
using
history
of
America's
is
more
of
a
partier,
but
my
experience
has
been
that
working
with
kerberos
directly
and
say
p
eyes
directly
in
mi,
tÃo
or
heimdal,
which
I
guess,
and
I
think
I'm
all
sort
of
dying
or
even
dead,
now
program
default
would
agree
with
that.
But
I
think
I
think
I
think
it
isn't
is
this.
L
Is
it
much
more
explicit
and
direct
and
clear
access
to
the
perimeters?
You
know
what
they
are.
This
is
AP.
I
was
opposed
to
support
a
lot
of
authentication
capabilities.
You
know
our
strategies
with
different
different
capabilities
and
a
sect,
and
essentially
only
the
holy
kerberos
really
exists.
What
I'm
from
ruthless
by
electron
happen?
Do
you
hear
that
I
hear
I,
don't
look
but
but
a
bit
of
a
do.
It
would
do
it,
but
if
we're
not
going
to
submit
up
an
approximately
a
kerberos
environment,
they
might
do
something
else.
No,
that's
that's!
That's!
G
L
I
mean
some
parts
are
some
bugs
work
very
well
and
are
well
understood.
You
know
we're
working,
working
with
principles
and
and
attributes,
and
so
forth
are
very
rich
working
with,
for
example,
encryption
and
not
as
much
and
their
incursion
is
supported
in
NFS
has
been,
in
my
view,
hindered
by
by
gssapi
support
rather
than
helped.
So
it's
just
now
sort
of
just
the
last
couple
of
years,
but
become
possible
to
do
more
interesting
stuff
and
dude.
Would
you
to
work
by
okay?
I
really
like
it's
the
microsoft
immunity,
mostly
but
yeah,.
G
L
H
One
of
the
things
that
I
was
concerned
about
when
I
you
started
right:
dude
sages
in
a
vmail
that
was,
if
we're
gonna
go
for
her
girls,
both
likely
going
to
rely
on
Elda
and
then
now
we're
lying
on
to
external
services.
To
do
depending
and
I
wonder
how
much
dependence
you
wanna,
we
wanna
wait
for
that.
You
know
we
have
two
rules
and
then
inhale
that
dissolve
we're
not
going
to
do
much.
H
B
Think
this
is
pretty
much
why
we
we
implemented
our
own
or
one
of
the
one
of
the
reasons
why
we
live
in
our
own
suffix,
instead
of
just
using
Kerberos
from
the
get-go.
But
I
think
I
think
the
way
this
would
actually
be
implemented
is
we
would
define
a
new
off
type.
We
already
have
none
and
SEF
x.
We
would
define
a
new
one
called
curb
five
or
whatever
it
is.
B
I
would
interact
directly
with
a
lib
Chris
library
which
seems
to
eat
Matt's
preferred
rat,
at
least
if
10
line
we
decided
that
we
would
prefer
to
use
gssapi.
Then
we
define
another
authtype,
that's
called
gssapi,
and
that
would
also
work,
presumably
only
almost
as
well
they're
almost
the
same
drive.
It
just
use
a
different
set
of
libraries,
but
if
that
would
allow
you
to
use
kerberos
for
users
for
logging
into
the
system
to
administer
it.
B
So
if
you're
on
the
CLI-
and
you
want
to
do
something
then
and
you're
authenticated-
be
a
Kerberos
and
you
could
just
use
your
Kerberos
credentials
to
authenticate
to
the
cluster
and
then
do
whatever
you
want
to
do.
But
that
doesn't
necessarily
mean
that
the
demons
have
to
use
kerberos
authentication
still
also
define
with
the
same
roles,
SFX
keys
and
if
curb
presses
down,
you
could
authenticate
with
a
client
admin
key.
They
have
stashed
somewhere,
that's
using
setbacks
or
something
like
that
right.
B
B
So
I
think
that
the
steps
in
my
mind
would
be
inicia
to
find
the
new
authtype
with
kerberos
5.
If
the
client
negotiates
that
with
this
antique
with
monitor,
then
it
would
use
the
Kerberos
library
to
get
get
a
ticket
dope.
I
still
think
this
risk.
This
initial
step
of
using
would
curb
of
you
but
having
the
user
authentication.
These
are
gets
the
Kerberos
ticket
and
I.
Think
then.
Ok,
so
the
monitor.
B
It
can
give
it
a
ticket
that
could
be
a
suffix
ticket
so
that
the
rest
of
class
are
still
using
sex.
That
it
seems
like
a
sort
of
a
simple
way
to
get
sort
of
the
usability
story
and
without
having
to
do
all
the
demon
rotating
key
stuff.
I'm
using
Kerberos
primitives,
which
I
think
is
going
to
be
a
lot
more
work.
K
L
C
A
G
M
It
we've
been
ed,
I'll,
say:
I
regret
it,
but
it
thank
you
for
telling
me
Apple
moved
on
that'll.
Give
me
more
ammunition,
try
and
get
rid
of
it,
because,
oh
what
a
mess.
B
B
G
B
G
I
think
what
you
want
to
do
is
is
set
up
a
mechanism
where
most
Kerberos
users
don't
have
an
explicit
identity
in
the
South,
have
user
database
and
and
somehow
they
get
mapped
in
in
some
some
form.
You
can
have
posix
attributes
out
in
in
ldap,
and
that's
probably
the
most
natural
model
for
how
you
would
want
to
interact
with
that
and
Seth.
Well.
L
Yeah,
if
this
get
this,
as
this
is
done
for
was
bad
but
I
mean
the
idea
of
paca.
Sawyer
to
this
I
mean
we
Daniel's,
mentioning
all
that
/
things
before
earlier.
The
first
the
most
can
be
a
way
to
think
about
it.
To
me,
it
seems
like
the
most
convenient
way
to
think
about.
Things
is
to
not
include
not
worried
about
that,
but
but
you
know,
but
almost
everything
that
has
dinner.
L
You
know
that
I
under
design
for
their
with
DCFS
dce
and
Microsoft
processed
I,
was
in
net,
where
eventual
avenge
eventually
come
up
with
a
bit
more
or
less
agree
on
this
more
or
less
this
mapping.
But
the
purpose
wisdom
was
to
was
to
mr.
quest
to
create
that
was
supposed
to
beat,
was
to
create
the
domain
database
for
all
the
objects
in
the
end
there,
and
then
the
environment
and
organize
it
there
and
kerberos
design,
as
Marcus
has
just
map
to
it.
L
But
that
means
that
you
yeah
no
one
and
you
it
means
means
we
can
deal
with
it
later.
But
what
was
we
do
decide
to
deal
with
it?
Then
you
have
to
well
define
things,
and
this
is
I
try
to
be
in
some
on
the
Pat
and
the
pack
discussion.
You
really
can't
talk
about
pack
as
given
without
talking
about
ad.
It
basically
basically
contains
ad
binary
information.
L
Now
the
idea
has
been
generalized,
and
you
know
the
kerberos
community
intends
to
do
so,
but
they
don't
have
an
implementation,
but
the
idea
is
you
enhance
your
KDC
to
propagate
the
group,
information
or
other
other
other
other
other
other
secret
ticket
information
abstractly.
However,
you
wish
so
a
super
KDC
could
do
that,
but
that's,
but
that's
quite
a
way.
That's
why
I
called
it
blue
skies
wasn't,
president
doesn't
exist
today.
So
it's
not
something
we
designed.
J
J
B
Like
chef
manager,
rule
I
think
that
the
authenticating
clients
is
that
sort
of
critical
step,
and
that
happens
in
order
to
bootstrap
their
access
to
the
cluster,
so
I
think
I
think
it
belongs
on
the
monitor
and
in
fact
you
generally
wouldn't
even
tell
clients
who
the
managers
are
until
they
authenticated
and
the
authentication
key
said.
You're
allowed
to
talk
to
you,
these
other
people.
J
B
B
J
L
N
L
Begin
to
get
here
because
I
mean
the
business
is
just
a
business
gets
into
how
we
advanced
users
of
how
we
want
to
define
Kerberos
protocols.
The
simplest
models
don't
require
that
at
all
the
KDC
can
get
the
service
tickets.
You
need,
and
every.
G
Ya
self,
it's
kind
of
interesting,
because
you've
got
multiple
different
pieces
of
it
that
won't
talk
to
each
other,
and
then
you
have
users
in
the
mix
and
so
what
you're
actually
probably
kind
of
looking
at
there
is
some
sort
of
proxy
arrangement
where
you
is
user,
a
acting
to
server
X,
which
is
then
talking
to
surfer.
Why
and
server?
Why
would
probably
see
your
identity
is
kind
of
a
composite
of
server
X
and
your
identity.
L
G
L
B
L
But
they
went
further
than
that,
like
it's
like
we
service
to
commercial.
Rogue
servers
tickets
would
give
you
that,
but
but
you
would
believe
in
it,
but
in
this
a
RX
TK
model
you
would
you
initiate
with
it
with,
for
example,
client
wouldn't
issue.
Quite
a
cache
manager
in
traditional
AFS
client
would
have
never
tip
rentals
for
the
users.
Users
would
initiate
with
it
with
the
width
up
with
a
couple
with
an
initial
service.
L
B
So
if
s
clients
mount
points,
are
going
to
do
that
eventually,
because
you're
going
to
have
a
server,
that's
mounting
on
files,
nothing
and
then
multiple
users,
authenticating
on
that
server
and
the
client
is
going
to
be
practicing
those
identities
to
the
metadata
server.
The
OSD
is
in
order
to
do
whatever.
G
The
Mon
is
probably
a
simpler
case,
because
that
I
think
you
basically
have
kind
of
two
classes
of
things
that
talk
to
it.
One
of
them
is
you
have
administrators,
doing
administrative
actions
on
it
and
the
other
one
is
you
have
machines
who
are
just
basically
getting
basically
they're
using
this
kind
of
location
broker,
finding
out
where
resources
are
am.
B
Well,
I,
don't
I
don't
want
it.
I
mean
every
Kerberos.
Discussion
can
like
go
off
into
the
box
with
all
the
stuff,
so
I
want
it.
I
want
to
focus
on
sort
of
what
the
first
steps
are
to
actually
move
forward
on
this.
So
so
in
this
first
section
on
the
eighth
of
bad
goal,
one
there's
implementation:
this
is
this:
can
we
sort
of
agree
that
this
is
the
right
set
of
four
steps
and
Daniel
doing
it?
Is
there
any
business
this
make
sense
to
you
or
other
I.
I
B
G
It's
okay,
another
possible
variation
would
be
to
have
each
bond
have
its
own
host
identity,
identity.
I
know.
Sage
was
not
very
keen
on
that,
because
that
makes
the
provisioning
process
for
setting
up
a
mod
a
lot
more
complicated,
but
it
is
another
viable
model.
I
had
only
interesting
thing,
I
see
here,
yeah.
B
H
L
B
B
N
Cool
one
other
thing
that
cursed
me
here,
we
sort
of
occasionally
talk
about
in.
N
Oh,
we
want
it
to
be
not
just
Kerberos
in
the
future
and
just
sort
of
the
motivating
one,
because
that's
the
first
one
and
I
just
want
to
bring
that
up
again,
because
I
was
thinking
about
Keystone,
which
I'm
sure
we
all
hate,
but
also
we're
going
to
be
in
a
lot
of
OpenStack
times
and
eventually
not
now,
but
eventually
we're
going
to
need
what
we
built
here
to
at
least
be
a
stepping
stone
to
integrating
a
few
stone
for
user
identities.
N
B
N
J
H
B
Think
that
the
tricky
pit
bit
here
is
that,
where
you're
going
to
start
out
with
a
Kuro
stickit
and
then
once
you
got
that
Nikki
with
the
monitor,
then
it's
going
to
sort
of
switch
to
like
when
using
the
internal
off
method.
I
think
that's
the
weirdness,
that's
going
to
happen.
It's
going
to
be
annoying
and
we'll
deal
with
it
when
you
sort
of
stumble
into
the
details,
but
once
that
solved
then
I
think
Keystone
will
be
easy,
because
we
want
to
solve
the
hopper
and
that's
just
spotting.
It
then.
B
Well,
I
think
that
the
nice
thing
is
that,
even
if
we
don't
solve
that
that
weirdness,
though
it's
still
useful
because
you
can,
if
even
if
you
just
solve
the
client
authenticating
to
the
monitor
and
you
ignore
it,
giving
you
a
ticket
that
lets
you
talk
to
other
demons,
you're
still
extending
it.
The
monitor,
so
you
can
deal,
do
still
do
like
ninety
percent
of
the
administrative
commands,
and
so
you
have
something
that
we
can
test
and
we'll
be
moderately
useful.
B
You
just
won't
be
able
to
like
Oh
s,
T
tau,
I
guess
so
so
I
can
go
from
there.
B
Please
I
don't
know
if
I
can
decline
to
snapping
directly
I
mean
I
guess
you
could
have
some
other
extraneous
service
that
use
kerberos
to
like
get
a
suffix
ticket,
and
then
you
just
do
everything
else
the
same
the
same
way,
but
I
think
the
idea
is
to
try
to
get
eventually
sort
of
face.
Fx
out
of
the
picture
at
least
I
would
sleep
better
if
eventually
we're
wheezing
knots,
FX,
that's
something
but
I
think
it's
gonna
take
a
while
to
get
there
so
yeah
anyway.
B
All
right,
this
we're
still
no
afraid.
Oh
ok,
let's
go
to
Bassam
runtime
detection
of
cindy
instructions
for
the
EC
libraries
yep.
J
Should
be
straightforward
and
hoping
that
so
essentially
today
we
have
all
these
different
flavors
of
plugins
for
all
the
different
instruction
sets
thanks
the
ones
we
have
are
a
generic
one
for
Jerry
sure,
there's
an
fec
three
flavor
and
FC
for
flavor,
and
then
on
the
arm
architecture.
We
have
a
generic
mom
and
we
have
a
neon
one
which-
and
you
know
you
can
imagine
that
I
actually
just
by
looking
at
what
we
have
right
now-
we're
missing
or
missing
one
there's,
a
careless
multiplying.
J
That
is
missing
because
we
don't
actually
create
a
flavored
word,
but
and
if
I
think
it
actually
has
a
quite
a
bit
of
improvement
in
performance
if
we
were
to
enable
it.
But
it's
got
me
thinking
unto
why
why
do
we
need
to
support
all
these
different
flavors
and
not?
You
know
the
what's
in
style
now?
J
Is
you
build
a
library
once
and
it
because
the
runtime
detection
of
instruction
sets
and
runs
the
appropriate
code
is
needed
and
where
I
noticed
we're
doing
that
with
a
new
compression
stuff,
I
g11
is
a
stuff
that
was
just
committed.
So
this
is
an
attempt
to
do
a
similar
thing
with
the
ratio
coating.
J
You
know
in
a
nutshell,
you
built
you
build
Cherisher
and
GF
complete
and
once
if
with
a
compiler,
that
has
everything
turned
on
so
cross,
compiling
is
possible
as
well
and
then
at
run
time
when
they
hit
time
of
GF,
complete
and
Jerry
sure
we
detect
the
processor
and
look
at
what
simply
instructions
are
possible.
It's
a
really
cheap
operation
and
Intel
platform,
it's
just
looking
at
cpuid
and
then
the
binary
itself
as
all
the
different
flavors
compiled
in
that
we
just
run
the
appropriate
one
yep.
J
So
it's
a
very
significant
improvement,
I
think
to
kind
of
provisioning
and
trying
to
figure
out
which
library
to
use
and
I
you
know
look
but
in
life
in
all
likely
of
people
that
are
doing
this
right.
Are
writing
the
same
detection
code
at
runtime
and
some
animal
script
or
something
yeah.
B
C
B
C
B
G
B
Compile
a
bunch
of
different
plugins
with
all
the
different
various
flags,
so
there's
like
Jerry
sure
sis
you
follow
yeah.
This
is
like
the
simplest
thing
would
be
to
big
make
just
like
the
Jerry
sure,
auto
that,
like
figures
out
texture,
processor
and
then
switches
to
whichever
pile
bearing
is
the
right
one
or
you
could
have
one
that
has
like
if
stephs
or
I
don't
know
whatever
dynamic
branching
and
all
the
code.
I,
don't
know
how
you,
whatever
it
just
a.
C
You'd
really
optimization,
you
need
to
be
very
careful
about
that,
and
so
that
that's
what
I
was
going
to
point
out.
So
the
problem
right
now
is
that
we
have
different
racial
coding
profiles
that
are
mathematically
identical
but
have
different
names
because
they
require
different
instruction
sets.
That's
the
problem
right.
C
So
the
first
thing
we
want
to
do
is
identify
which
ones
are
supposed
to
be
identical
and
build
test
to
make
sure
that
they
really
are
then
yeah.
Once
your
code
exists,
that
dynamically
switches
or
I
mean
honestly,
you
could
put
it
in
the
config
file
at
first.
It
doesn't
matter
that
I
like
it
could
be
in
the
OSD
config
file,
which
variant
it
should
use.
C
But
the
important
thing
should
be
that
the
OSD
map
shouldn't
shouldn't
specify
the
OSD
map
that
has
the
pool
information
needs
to
deduplicate,
all
of
the
various
jayesh
or
whatever
very
cozy,
and
what
have
you
down
to
the
same
name,
and
it
needs
to
do
it
transparently
behind
the
scenes
without
requiring
users
to
do
anything
yeah.
That's
what
I'm
p.
J
From
what
I
can
tell
so
far,
the
maturity
I've
already
done
the
bulk
of
the
changes
in
GF
complete
tonight.
The
merge
requests
from
what
I
can
tell
majority
if
the
work
is
should
be
transparent
to
whatever
the
algorithms
being
used
as
part
of
Jerry
sure,
because
it's
really,
what
are
you
doing?
No,
you
know
that's
not.
C
The
case,
no,
what
I'm
saying
is
the
irritating
part
is
going
to
be
fixing
our
sins
right.
Now
we
write
into
the
OSD
map
like
what
race
your
clip
profile
of
pools
supposed
to
use.
G
C
Need
to
transform
whatever
is
currently
scribbled
in
there
to
the
correct
instruction
set
stripped
variant.
Basically,
you
need
to
have
every
possible
version
of
co
sheet
onto
one
common
version,
transparently,
without
people
requiring
it
without
requiring
people
to
do
so,
because
there's
no
way
we're
going
to
be
able
to
explain
how
to
do
this
properly.
That's
impossible
map
it
onto
the
generic
thing
as
an
interesting
legit
exactly
and
then
there
will
no
longer
be
any
of
those
other
variants.
So
that's
that's
going
to
be
a
bit
of
a
trick.
J
J
B
That's
what
I'm
saying
I
should
say:
oh,
let's
just
say,
j,
richer,
Jerry,
yeah
yeah,
exactly
that's
basically
Sam
saying
so.
If
you
say
Jerry,
Shore
sse4,
it
actually
means
Jerry
sure
that
then
auto
detects
whatever
you
have
and
then
does
whatever
your
processor
can
do,
I
think
that's.
Currently
there
will
just
become
meaningless.
Basically,
we.
C
Could
have
that
actually
error
out
and
just
say
that
is
not
go
name
of
a
plugin
anymore
I.
Don't
think
we
should,
because
there
are
existing
tools,
but
insofar
as
we
bother
to
not
error
out
on
those
it's
just
to
make
other
tools
not
break,
we
want
as
much
cause
we'll
get
rid
of
even
the
ability
to
think
about
the
instruction
set
that
the
algorithm
is
implemented
up
but
could
tell
Jerry.
M
C
Right,
so
what
I'm?
What
I'm
saying?
Is
you?
Don't
you
don't
wipe
them
out
but
as
if
you
decode,
that
is,
if
you
actually
pull
it,
OS
dmap,
binary,
encoding
off
disc
and
Dakota
the
new
code,
the
enlightened
code
should
just
transparently
map
it
back
to
the
non
instructions
that
one
because
the
enlightened
versions,
the
code
couldn't
possibly
care
less
right.
So
I
guess.
C
J
M
C
B
That
in
crackin
and
or
luminous,
basically,
when
we
have
an
OST
map
that
says
Jerry
sure
underscore
something
sse4
sse3
whatever
it
is,
it
just
interprets
it
as
Jerry
sure
violently
and
then
what
continues
to
be
a
ranger
plug-in,
auto
detects
and
uses
whatever
things
are
there
and
also
on
the
monitor.
If
we
see
an
OSD
map
that
has
Jerry
sure
underscore
anything,
it
generates
a
health
warning
that
says,
by
the
way,
your
pool
is
specifying
a
processor,
specific,
plug-in
name.
C
C
C
M
C
B
C
M
B
B
Next
up
shots,
we
still
have
alfredo
right,
yeah,
so
I
put
messenger
to
plans
on
here,
just
to
figure
out
what
the
steps
forward
are.
I
think
that's
conversation
actually
want
to
have
with
how
am
I,
though
I
think
he's
not
here.
So
I'll
just
do
it
on
the
list.
I
think
we
can
defer
that
to
later,
and
so
the
last
thing
on
there
is
really
the
Alfredo's
plants
were
set,
deploy,
set,
danceable
and
stuff
installer
and
he's
not
here
right,
I'm,
not
just
missing
him
on
the
list.
K
B
Let's
see
Virata
is
suggesting:
do
we
talk
about
some
of
the
new
encoding
and
decoding
stuff?
I'm
Marty
I
had
a
couple
glasses
of
wine,
so
I
don't
know
if
I'll
be
super
useful
in
this
discussion.
I'm
already
super
confused
by
all
the
different
options
here,
but
we
do
have
everybody
who's
sort
of
dip
their
toes
in
here.
So
I've
copied
you
on
that
last
email,
Sam
I,
think
that
we
have
sort
of
two
I,
think
I,
think
they're,
two
orthogonal.
B
So
there's
there's
one
thing
which
is
adding
this
idea
of
estimate
that
your
branch
does
stamps.
Are
you
still
haven't
code
as
exist?
You
still
decoded.
C
B
C
That
up,
because
it
looked
like
that
it
and
figured
out
so
I,
don't
care
but
ya,
know
that's
the
right
way
to
go.
If
you
do
go
the
antec
route,
what
you
there
would
be
an
instance
of
that
template
that
notices
and
generates
out
the
full
traits
thing
from
it.
That's
how
it
work.
You
want
the
ability
to
generate
a
separate
estimate,
one
so
the
core
one
should
be
the
trainer
of.
C
B
B
Okay,
so
that's
I'm.
This
is
the
part
that
I
get
lost
on
so
these.
So
there's
this
traits
object
that
defines
these
methods
for
a
given
type,
and
it
is
the
idea
that,
if
you
don't
explicitly
define
it,
then
you
would
because
I
that
defines
a
traits
method
called
encode
that
encodes
type
T,
but
right
now
what
we
actually
want
is
sort
of
a
top-level
and
code
function
that
takes
a
tea
argument.
Yes,.
C
B
C
O
C
A
look
at
it
tomorrow
on
the
main
trick.
I
wanted
to
show
you
guys
was
I
I,
don't
want
you
using
my
branch
exactly
because
it's
not
as
like
it
used
the
sages
offender
and
the
consensus.
Is
we
don't
wanna
go
that
direction
anyway?
I
mostly
wanted
to
introduce
the
trick
that,
in
that
one
in
integrates
the
features
stuff
properly
so
that
it
automatically
does
that
for
ya,
especially
for
pairs
and
maps,
and
that's
the
main
trick
I
want
to
get
in
there.
C
I
want
this
to
work
for
the
existing
existing
types,
so
I
wanted
to
hook
into
the
current
system.
I
think
these
are
the
one.
B
Sure
you
guys
know
better
the
other.
The
other
key
thing
that
I
would
like
to
get
in
there
is
if
we
can
define
an
encode
in
terms
of
a
template,
so
that
appender
could
be
whatever
I
think
that
actually
is
the
right
route,
because
we
could
actually
define
that
a
tent
that
appender
type
could
actually
just
be
a
car
star
p,
and
you
have
a
couple.
You
two
types
that
and
we
could
actually
use
it.
That
way
too.
So.
O
The
end
code
function
does
try
to
do
the
unsafe,
sharpener
and
safe
appender.
We
could
replace
with
the
casual
fee
if
we
get
the
courage
to
make
there.
So
it's
still
a
templated
okay.
B
I
O
Yeah
so
so
there
are
some
map,
the
containers,
especially
the
maps,
some
of
the
the
first
one
has
new
features
like
streams,
and
then
some
of
them
has
features.
So
that's
where
I
kind
of
I
don't
know
what
to
do
there,
but
maybe
I
will
just
hope.
I
don't
really
have
this
one.
The
features
right
now.
I
think.
B
B
It,
and
if
neither
of
them
do,
then
you
don't
need
to
but
the,
but
if
you
don't
have
to
deal
with
that
case,
because
is
you
can
always
pass
features
to
an
encode
function
that
doesn't
take
features
it'll
just
ignore
it,
I
call
the
encodes
I
think
are
defined
in
terms
oh
and
actually
they
aren't
now.
B
Exactly
right,
so
the
ones
that
don't
require
features
are
at
code
that
top
of
own
code
function
as
takes
a
default
argument.
It
sets
it
to
zero,
so
you
can
always
pass
it
in
will
just
get
ignored,
but
the
ones
that
actually
do
require
their
features.
Don't
have
a
root
yeah,
so
I
think
that
I
put
it
an
email.
Basically,
if
either
of
them
request
features
than
the
gel
type
in
requires
features.
Yeah.
B
O
C
C
I
C
C
O
B
B
We
killed
I
don't
make
today.
I
was
very
planning
big
thanks
to
Holly
and
Casey
and
didn't
keeps
moving
everyone
who
do
a
lot
of
heavy
lifting
there.
Yeah.
C
Long
until
jewel
is
deprecated
yeah.
A
Cool
all
right.
Well,
if
that's
everything
then
just
put
a
reminder
in
here
for
the
video
for
people
watching
later
that
the
the
next
def
developer
monthly
will
be
on
Wednesday
October.
The
5th
will
be
back
on
a
stern
and
eastern
time
zone
friendly,
1230pm
or
rmaf
friends,
and
those
of
us
who
want
to
have
a
normal
sleep
schedule.
So
we'll
see
you
guys
back
on
the
5th
and
then
hopefully
everyone
is
going
to
OpenStack
Barcelona.
If
you
haven't
purchased
your
tickets
yet
check
the
lists.
I
sent
out
a
discount
code.