►
From YouTube: Ceph Orchestrator 2022-12-06
Description
Join us weekly for the Ceph Orchestrator meeting: https://ceph.io/en/community/meetups
Ceph website: https://ceph.io
Ceph blog: https://ceph.io/en/news/blog/
Contribute to Ceph: https://ceph.io/en/developers/contrib...
What is Ceph: https://ceph.io/en/discover/
A
Hey,
let
me
get
started,
I
know,
there's
a
holiday
in
Spain
today,
oh
right
over
here.
A
Hey
I,
I
fear
the
beat
probably
two
topics
today,
one
that
I've
already
put
into
the
other
pad,
which
is
I
think
from
openstack.
Let's
go
talk
about
be
a
place,
location,
tracker
and
then
I
figure
we'll
probably
end
up
talking
about
the
NFS
stuff
as
well
a
little
bit
so
some
people
are
are
in
here
who
know
about
that.
C
Hey
yeah,
so
I
I
just
created
this
tracker
to
compare
what
we
did
with
the
fancy
ball
in
the
past
and
what
is
done
today
by
7dm.
So
it's
not
high
priority,
which
means
we
don't
have
to
rush
on
this
kind
of
change.
But
the
the
main
topic
here
is
that
with
safangible
we
were
able
to
customize
the
crash
device
class
per
OSD,
while
safety
M
just
use
safe
volume
in
batch
modes,
you
pass
the
list
of
osts
and
you
just
prepare
and
activate
them.
C
So
this
is
mostly
for
our
CI,
where
we
kind
of
bake
the
ssds
to
create
crash
hierarchies
and
apply
rules
on
just
devices
that
are
all
the
same,
but
we
need
to
assign
sign
a
different
crash
device
class
in
order
to
be
able
to
make
sure
we
have
coverage
for
that
kind
of
feature.
C
So
that's
that's
the
thing
and
I
created
this
jogger
and
I'm
wondering
if
we
want
to
have
different
Syntax
for
the
device
class,
the
the
path
that
we
pass
to
the
drive
group,
or
we
can
just
make
that
part
smarter
enough
to
process
device
classes
per
OSD.
C
A
A
C
Oh
sorry,
I
didn't
realize
this
is
the
old
one
and
basically,
what
it
does
is
setting
the
crash
device
class
as
a
global
parameter,
which
means
that
you
define
a
list
of
osds
and
that
crash
device
class
is
applied
to
all
the
devices
because
you
run
safe
volume
in
batch
mode.
So
you
pass
this
list,
this
big
list
of
osds
and
you
apply
this
device
class
to
all
of
them.
While
my
the
tracker
High
opened-
and
let
me
and
it's
I,
can
it's
in
the
other
part
here.
B
A
I
mean
if
it
was
something
we
could
do
in
San,
Francisco
I
have
no
problem
with
having
it
here.
I
guess.
The
the
only
question
is
is
how
we
want
to
format
the
yaml
here.
C
Yeah
I'm
not
sure
how
much
popular
was
this
picture
that
we
hadn't.
A
C
A
Heard
about
it
so
I
don't
think
a
lot
of
people
were
using
it,
but
definitely.
C
Yeah,
this
basically
allows
to
have
more
coverage
in
TI,
so
as
a
side
effect
of
this
missing
feature,
we
are
not
able
to
basically
duplicate
the
same
jobs
that
we
had
with
the
fancy
board
and
make
sure
they
worked
with
safe
ADM.
C
A
Yeah
so
I
have
no
problem,
that's
just
functionality-wise!
It's
just
a
matter
of
how
we
want
to
format
this
in
the
emoj.
This
is
why
I
asked
you
you
wanted
to
bring
it
up
in.
Do
you
want
to
discuss
that?
Because,
obviously
we
have
this
thing
where
right
now,
the
path
section
is
just
a
list,
and
so
we'd
have
to
have
this.
This
section
in
here
where
we
do
a
crush
device
class
under
that
yeah.
C
I
tried
the
first
PLC
on
my
Dev
environment
and
basically
you
can
just
make
this
list
a
dictionary
using
this
format
described
in
the
in
the
tracker
and
you
can
process
you
can
create
this
device
object.
C
There
is
the
device
class
python
within
drag
groups,
and
you
can
add
this
additional
Fields,
the
crash
device
class,
so
it's
assigned
per
device
and
then
there
is
a
device
selection
in
the
drag
groups
that
you
can
use
to
filter
your
devices
that
would
be
prepared
by
set
volume
and
basically,
what
I
found
is
that
you
have
to.
If,
if
there
is
this
kind
of
syntax
in
the
spec,
you
can
process
them
in
a
different
way,
which
means
you
have
to
create.
C
You
can
still
rely
in
the
batch
mode
for
set
volume,
but
it's
more
the
interactive
mode
here,
where
you
prepare
each
device
like
in
a
for
Loop
one
by
one
and
you
were
able
to
pass
the
device
class.
So
the
the
pattern.
The
approach
is
a
bit
different
because
in
batch
mode
you
have
the
whole
list
and
you
have
the
crash
device
class
as
a
global
parameter
applied
to
the
whole
list.
Well,
here
you
should
create
this
Loop,
where
you're
creating
you're
running
it's
a
volume.
C
A
All
right:
well,
it
sounds
like
at
least
the
the
ammo
private
won't
be
that
bad.
If
it's
just
the
native
dictionary
it
just
works
for
there,
yeah
I
guess
we'll
have
to
figure
out
the
batch
things
that
will
slow
down
the
OSD
deployment
a
lot.
A
If
you
have
to
do
them
all
individually
for
that,
unless
we
can
I
guess
study,
what
we
could
do
is
maybe
we
could
sort
them
by
the
ones,
because
it
seems
like
it's
mostly
just
SSD
and
HDD
here,
so
maybe
we
could
sort
them
by,
like
all
the
ones
with
SSD.
We
can
put
it
in
one
batch
or
something
like
that.
I'm
not
sure.
C
Yeah,
you
can
basically
optimize
this
creation
by
grouping
disks
by
crash
device
class
and
then
run
the
set
volume
in
batch
mode,
applying
that
that
label
to
hold
this,
that
you
define
in
the
spec.
So
basically
this
could
optimize
a
bit
the
way
you
apply
this
Crush
device
class,
because
it's
true
that
using
this
syntax,
you
are
able
to
apply
this
label
per
OSD
per
device,
let's
say,
but
then
you
can
build
two
or
at
least
three
lists,
and
you
have
to
run
two
or
at
least
three
set
volume
commands.
A
Right
we'll
do
that
I
guess
we'll
have
to
also
be
careful
about
the
default
one,
because
we
still
have
obviously
the
spec
wide
option
as
well.
So
it's
about
to
be
set
as
like
a
default
like
I
guess
if
you
said
pass
and
that
one
we
have
to
have
the
path
be
the
more
important
one
and
then
have
this.
The
spec
wide
one
be
the
default
and
we'll
also
have
to
I
guess
daily
counter.
Some
of
them
are
going
to
have
no
device
classes.
A
A
That
doesn't
sound
like
it'll,
be
too
bad,
at
least
if
it's
only
three
batch
commands
per
spec
at
most
I
mean
that's
only
people
that
you
want
to
use
this
and
want
to
set
a
bunch
of
different
devices
which
again
I
think
it's
not
a
super
widespread
use
case
for
what
I
can
see,
but
but
the
performance
of
most
people
shouldn't
be
affected
and
then
I
guess
you'll,
maybe
see
it
a
little
bits
if
you're
doing
this
with
like
a
bunch
of
OCS
in
one
hopes,
but
it
shouldn't
be
too
bad.
A
That
sounds
good.
You
know,
I
think
that
sounds
reasonable,
that
we
will
use
a
dictionary
in
the
spec
and
then
we'll
set
up
those
three
up
to
three
things:
we'll
have
a
little
bit
of
logic
that
sorts
them
through
the
spec
and
everything,
but
I,
don't
think
it
sounds
too
bad
gone
through
it.
A
I'll
probably
update
the
track
or
whatever,
with
some
of
the
stuff
we
talked
about
as
well,
but
that
sounds
good
anything
else
you
can
say
about
it.
This
is
like
this:
I
have
to
go
with
budget
now.
C
Yeah
thanks
I
agree,
so
we
can
best
basically
group
the
disks
and
run
up
to
three
step
volume
commands
and
that's
good
enough.
A
A
That
covers
that
topic:
I,
don't
have
it
in
the
agenda
as
well,
but
I
know
people
are
already
talking
in
the
chat
about
the
NFS
stuff.
B
Yes,
so
it
started
out
with
actual
debugging
and
then
turned
into
me:
debugging
the
fact
that
I
can't
apparently
send
messages
to
IRC,
even
though
I
can
read
the
conversation.
B
So
it
sounds
like
there's
some
potential
debugging
in
RC
Frank
is
saying
that
he
may
have
introduced
a
regression
with
a
new
config
section
and
then
over
to
chat.
Romana
is
suggesting
how
I
can
add
it
to
the
config,
and
that
is
you're
getting
like
three
seconds
ago.
B
That's
how
up-to-date
you
are
so
yeah,
we'll
I'll
try
some
of
these
things
after
the
meeting
and
yeah
and
then
and
then
long
after
the
meeting
I'll
try
and
figure
out
why
I
can't
use
IRC
correctly,
but
that's
different
yeah.
It
sounds
like
this
could
be
the
culprit,
so
we'll
we'll
see
if
it.
If
the
behavior
changes
with
this
config
section
or
not,.
A
All
right
cool
sounds
like
progress:
therapist,
yeah,
I'll
copy,
the
IRC
conversation
probably
into
the
other
path.
I
know
some
people
either
missed
it
entirely
or
joined
part
way
through
or
something,
and
you
can't
go
back
and
get
those
I
think
so.
A
I'll
put
that
in
I'll
put
it
up
to
here
at
least
I'll
copy
it
in
there
there's
not
really
after
you
discuss
that
much
in
here
it
sounds
like
it's
they're
getting
closer
to
finding
out.
What's
what
happened.
B
Yeah
I'm,
just
I'll,
probably
need
a
little
bit
more
hand-holding
because
I'm
not
I'm,
familiar
enough
with
the
format
from
hacking
on
the
python
side
of
the
module,
but
that's
modifying
what
was
already
there.
This
is
like
adding
a
new
config
block,
and
this
is
one
of
those
things
where
I
usually
have
like
documentation
on
in
my
browser
window
as
as
I
hack
on
things,
I,
don't
know
exactly
where
it
fits.
B
B
So
it
should
be
good
thanks
now,
you're
already
being
helpful
in
in
g-chat
I,
just
I'm
kind
of
catching
up
to
where
you
guys
already
are
far
ahead
of
me.
A
D
I
mean,
as
I
already
emailed
the
two
of
you
and
the
Manila
team
I'm
moving
over
to
the
RPD
team
to
help
them
out
because
they're
short
staffed.
So
please
reach
out
to
the
self-esteem
and-
and
you
can
also,
of
course,
reach
out
to
Frank
Phils
who's
like
who's
the
PTL
of
the
Ganesha
project,
so
yeah
you'll
have
all
the
info.
You
need.
A
Yeah
Frank
Frank
should
be
a
good
contact
to
have.
We
have
to
reach
out
to
the
SFS
team
a
bit
earlier,
but
I
think
it
ended
up
back
at
you
like
it
went
from
I
asked
Daria
I
think
he
went
to
Greg
and
then
went
back
to
you
after
that.
D
Yeah
transition,
yeah
I,
think
it'll
take
a
while
for
him
to
catch
up.
You
know
until
then
yeah
Greg
wanted
me
to
help
you
guys
out.
Hopefully
they
can
do
that
quickly.
You
know
get
him
off.
Nobody.
A
Doesn't
come
up
too
often,
I
know
this
is
the
second
time
now.
This
has
happened
where
we've
had
some
big
problems
with
this
announcing
these
things,
but
I
think
last
time
it
resolved
itself.
I
came
up
and
on
our
side
with
some
the
way
we
were
kind
of
building
the
images,
but
at
least
this
time
we
have
you're
still
here
to
help
and
we'll
we'll
be
able
to
ask
Frank
I
guess
in
the
future.
A
This
comes
up
again.
So
that's
really
helpful.
B
A
Hi
I'm
not
gonna,
do
so
much
to
say
more
about
that
in
here,
something
that
it
has
to
be
handled.
Offline
people
are
looking
to
it.
That
should
be
the
last
issue
we
have
in
our
testing.
Speed.
I
saw
update
a
little
bit
on
that
I
did
a
testing
run
that
finished
a
couple
days
ago,
and
that
one
was
mostly
good.
A
Obviously
still
had
this
issue
in
there,
and
there
was
some
other
rados
issue
which
turned
out
to
be
a
actual
issue
that
got
merged
in
some
regression
got
merged
a
few
days
ago
or
a
while
back
actually
now
we're
just
listened
to
welcome
the
test
until
now,
but
other
than
that
our
run
was
pretty
good,
so
I
got
the
finally
able
to
start
merging
things
again.
So
obviously,
once
we
have
this
NFS
stuff
figured
out
they'll
be
down
to
only
a
handful
of
small
issues,
mostly
like
installing
packages
and
like
some
stuff.
A
That's
mirror
timing
out
like
that.
We
can
go
dress
directly,
but
we're
pretty
close
I
think
once
we
have
the
NFS
stuff
figured
out
to
being
back
to
having
a
pretty
clean
test
Suite
and
we
got
at
least
have
a
cleaning
up
now,
where
we
can
start
merging
things
again,
so
it
shouldn't
be
as
stuck
as
we
were
for
pretty
much
the
entirety
of
November
cool.
A
That's
really
good,
that's
all
I
had
for
today
does
anyone
else
have
any
other
topics
they
want
to
bring
up
in
here.
A
B
Is
going
to
follow
I'll,
give
more
details
in
the
PC,
but
I
believe
the
suggested.
Workaround
is
working
now,
oh
cool
good
here,
so
it
does
sound
like
there's
a
little
regression
in
Ganesha
when
this
option
is
not
present.
A
Well,
I
guess
we
can
handle
it
from
our
side,
at
least
they
can
work
on
something
there
afterwards.
B
A
A
Yeah,
we'll
do
that
after
we'll
figure
that
all
out,
at
least
this
gives
us
somewhere
to
start
with
and
know,
what's
sort
of
going
on
now
exactly.
B
A
Really
very
good.
All
right
in
that
case
have
a
good
rest
of
your
day.
I
guess,
I'll
see
you
next
week,
bye.