►
From YouTube: Ceph Orchestrator Meeting 2021-10-12
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Looking
at
it
again
now
we
have
one
two,
three
or
topic,
which
is
great
first
topic.
We
have.
A
The
entry
organization
has
started
its
winter
turn
and
I
have
added
the
number
integration
to
ceph
adm
and
I've
also
already
got
two
students
who
are
interested
in
the
project
and
yeah.
Let's
see
how
it
goes.
If
we
can
manage
to
get
some
more
integration
into
that
video,
it
would
be
really
awesome.
A
So
suffolk
upgrade
ls
sage.
I
guess
that's
you
right.
B
Yeah
I
was
taking
another
look
at
this
previously.
I
was
thinking
that
this
should
just
look
at
the
releases.yaml
file
that
we
maintain
in
the
stuff
repository
as
sort
of
a
canonical
the
list
of
like
all
the
versions
that
are
available
and
and
so
on,
and
we
could
even
like
mark,
which
ones
are
like
toxic
and
shouldn't
be
installed.
That
sort
of
thing.
B
If
we
have
like
a
release
that
we
want
to
pull
whatever
it
is,
but
it
occurred
to
me
that
that
maybe
isn't
so
good,
because,
if
you're,
using
like
a
custom
registry
locally
and
you're
pulling
them
just
from
somewhere
else,
you
probably
want
to
list
the
versions
that
are
available
there
there
and
not
the
ones
that
are
upstream
yep.
Possibly
so,
then
I
went
back
and
looked
at
your
pull
requests
and
it
was
doing
some
like
annoying
growth
stuff
to
look
at
the
registry.
B
And
so
then
I
look
to
try
to
find
if
there
is
a
python
library
for
interacting
with
the
docker
registry.
I
found
one
and
it
seemed
to
work
sort
of
with
clay,
but
I
couldn't
make
it
work
with
docker
and
it's
not
clear
that
it's
very
well
maintained
then
I
just
got
annoyed.
I
was
like:
why
isn't
there
just
like
a
library
for
dealing
with
container
registries
because
it
seems
like
that's
like
well,
the
only
thing
the
entire
world
wants
to
like
think
about
is
containers.
B
A
Supporting
always
migrating
to
the
latest
tag
using.
A
B
For
me,
I
would
like
to
be
able
to
like
see
what
versions
are
available
when
they
were
released
and
what
version
I'm
running
now
and
be
like.
Oh,
I
want
that
one,
that's
what
I
would
want
and
then
maybe
I
also
have
like
a
command
that
just
goes
to
the
latest
blindly,
but
I'm
not
sure
I
would
use
that
very
often.
B
I
think
the
thing
that
bothered
me
was
that
it
like
the
there's
one
string
that
that
you're
basically
telling
the
cluster.
This
is
where
I
want
to
get
my
painters
and
it's
going
to
be
like
way:
dot,
io,
that's
the
image
I
want
to
use
and
your
pull
request.
If
I
remember
correctly,
it's
looking
at
like
registry.hub.docker.com.
B
I
think
it's
still
hardcoded
at
that
time.
Right
yeah,
but
it's
not
docker.io
right
like
if
you,
if
you're,
specifying
an
image
on
docker,
have
it
stalker
dot,
io,
slash,
yes,
flash
image.
B
A
It's
only
docker
where
you
need
to
use
different.
The
dock
io
registry
has
different
sub
domains
or
different
parts
of
the
api.
Every
every
other
registry
is
same
okay,
but
what,
if
you're
deploying
like
the
docker
registry
on
premises,
then
everything
is
working.
Fine,
it's
just
the
docker
io
registry
that
is
using
those
change,
sub
domains.
A
A
Unreliable,
how
registries
might
choose
to
disable
listing
tags.
A
Not
being
able
to
rely
on
it
was
as
a
showstopper
at
this
time,
but
if
you're
just
using
it
to
display
upgrade
notification-
and
it's
perfectly
valid-
you
know,
it
is
just
that
we
can't
use
it
to
actually.
A
Create
a
same
cluster,
you
know
we
we
have
to
be
more,
we
we
can't
rely
on
it
in
order
to
function.
B
A
B
Yeah,
this
is
just
this
can
take
this
offline.
I
was
just
frustrated
because
the
rook
code
was
hitting
an
exception
and
the
rapper
was
basically
eating
the
trace
back,
and
so
you
can't
see
where
this
error
was
happening.
You
see
the
trace
back
for
where
that
gets
re-raised
later.
A
C
Hey
yeah,
we
if
you
recall
the
last
conversation
that
we
had
on
manila,
interacting
with
staff
and
cfs
with
ganesha.
C
C
So
we
came
up
with
a
spec
which
is
supposed
to
describe
our
scenarios,
the
use
cases
that
we're
trying
to
cover-
and
we
have
some
questions
that
are
that
have
some
implication
on
the
changes
that
we
need
to
make
within
the
manila
driver
to
interact
with
seth
and
ganesha,
and
that's
the
reason
of
this
this
topic
here.
C
The
the
first
thing
I
saw
is
that,
in
order
to
interact
with
ganesha,
we
have
to
create
an
export
or
delete
an
export.
We
have
some
cli
provided
by
the
orchestrator
and
this
there
we
have
this
commons
fnfs
export
or
delete.
We
can
get
info
from
the
cluster
or
described
your
exports.
C
We
can
do
a
lot
of
things,
so
the
first
question
that
came
out
was
if
there
is
a
way
to
interact
with
the
orchestrator,
not
just
not
only
via
via
dli,
but
also
using
libraries,
api,
and
you
know,
because
from
manila
we
can
define
the
share
and
then
translate
it
into
a
dev
common
that
is
consumed
by
an
api.
B
If
you're
just
talking
about
wanting
a
rest
api
as
opposed
to
a
cli,
then
the
sort
of
not
really
like
but
they're
the
cli
is,
is
not
just
like
a
cli.
You
can
also
feed
it.
You
can
use
the
orchard,
ply
and
feed
it.
The
yaml
specification
for
the
nfl
service.
You
can
do
it
that
way
and
one
of
those
nfs
commands
like
nfs
export
apply.
I
believe
it
is.
You
can
feed
either
a
adjacent
description
of
the
export
or
you
can
feed
it.
An
actual
ganesha
config
block
and
it'll
import,
those
yeah.
B
C
B
C
Yeah
so
in
theory,
we
can
start
investigating
this.
This
part
see
if
it's
better,
using
this
api
to
interact
with
the
orchestrator
or
just
use
the
cli
and
yeah
yeah.
D
B
B
Otherwise,
you'd
need
a,
I
guess,
the
manager
write
capability
or,
like
can't
remember
exactly
you
need.
You
need
to
write
capability
on
the
manager
in
order
to.
D
To
be
able
to
do
that
because
in
the
past
I
think
manila
was
going
to
top,
but
this
would
be
the
best
way
to
interact
with
the.
D
Okay
and
okay,
I
think
the
the
reason
why
I
was
asking
about
the
grants
is
that
in
the
past,
because
we
were
riding
into
the
pool,
we
didn't
need
specific
permissions
to
operate
against
the
cluster.
We
just
needed
permission
to
ride
into
the
pool.
So
now
that's
a
little
bit
more,
but.
B
B
Whatever
commands
that
you
end
up
using
it's
a
little
bit
tedious
to
write
those
those
capabilities,
but
it
is
certainly
possible
do
that
if
you
want
to
narrow
down
to
like
just
the
set
of
commands
that
you
specifically
need.
D
Okay-
and
this
is
how
do
we
do
that?
How
do
we
learn.
B
Yeah
I'll
need
a
minute
to
go
pick
it
up,
but
we
can
find
it.
I
had
a
question
about
this
this
piece
above
here.
C
Actually,
this
is
how
it
worked
in
the
past,
and
now
we
should
rely
on
this
new
interface
fnfs
interface
to
export
or
delete
a
share
without
the
need
of
changing
ganesha.com
for
reloading
it.
Actually,
my
real
question
was
related
to
updating
specific
portion
of
ganesha
conf,
because
I
see
actually
you
can
do
that
using
the
same
7fs
interface.
C
This
is
different
and
we
were
wondering
if
there
is
a
way
to
describe
how
your
nfs
demon
is
not
specific
time.
Maybe
we
can
ask
for
interface
that
generates
a
description
of
the
current
snapshot
of
ganesha
demon.
B
C
C
A
But
it's
only
one
user
config.
So
if
you're
overwriting
the
user
config,
then
you
can
actually
override
existing
blocks
yeah.
But
you
cannot
overwrite
blocks
that
are
written
in
the
fdm
ganesha
template,
so
the
the
nfs
module
user
config
is
restricted
to
things
that
are
not
yet
defined
in
the
ganesha.
That
confidently
too.
C
And
every
time
I
just
push
a
new
config
portion,
I
need
to
reset
the
entire
config.
As
far
as
I
can
see
from
any
experiment,
I
made.
A
C
Okay,
now
because
I,
for
instance,
if
I
just
apply
loads,
full
debug,
okay
in
my
nfs
cluster
and
then
according
to
the
document
session,
and
then
I
had
a
custom
export
block,
I
cannot
do
that
because
it
just
said
that
my
cluster
is
already
my
config
is
already
set.
So
I
need
to
reset
the
configuration
create
a
new
full
block
and
then
apply
it
again.
C
Yes,
if,
if
this
is
the
behavior
that
is
expected,
I'm
okay-
I
was
just
thinking
about
incremental
things
that
can
be
applied
against
organisations
without
the
need
of
reset
the
cluster.
The
user
config
in
the
middle.
C
A
C
The
first
thing
is
how
to
interact
with
this
component,
and
we
saw
that
probably
the
cli
is
the
best
option
that
we
have,
because
I'm
also
looking
at
the
nfs
ganesha
used
by
the
dashboard.
The
difference
between
the
dashboard
and
an
external
component
is
that
you
cannot
just
import
the
codes
that
reside
within
the
self
code
base.
A
Yeah,
the
the
cli
is
definitely
a
bit
more
direct
access
to
the
cluster
and
I
guess
what
you
want.
Otherwise,
you
will
inevitably
hand
up
to
patch
the
dashboard
in
order
to
make
it
to
extend
the
dashboard.
In
order
to
add
the
features
that
you
really
need
to
to
have.
A
If
you
want
to
have
fast
progress-
and
I
would
say
if
you
want
you,
you
could
look
into
avoiding
the
dashboard
at
this
point.
B
C
C
C
C
Yeah,
okay,
the
alternative
is
that
yeah,
because
the
alternative
is
run
several
comments
that
list
all
the
exports.
Then
the
info
against
all
the
for
each
export.
You
can
get
the
info
and
you
can
collect
all
the
information
connection,
fnfs
demons
and
then
you
can,
you
know,
build
the
description
of
the
f
in
their
gap.
B
Yeah,
I
think
that's
probably
I
mean
you
can
do
ls
like
export
ls
detailed,
I
think,
and
it'll
dump
all
the
exports
all
in
one
go,
but
that's
separate
from
the
like
the
cluster
info.
That's
probably
like
three
transits
like
cluster
info
dump,
all
the
exports
and
then
one
to
fetch
the
cluster
config,
which
we
would
have
to
add.
B
C
I
mean
I'm
not
sure
about
the
parameters
that
are
required
for
manila
for
the
new
changes
if
there
is
a
requirement
or
a
change
within
the
ganesha.com
templates,
that's
within
the
f
fadm
the
orchestrator,
but
at
the
moment
I
think
the
first
question
was
mostly
how
to
interact
with
this
new
command
and
if
it's
the
right
way
to
do
because
the
holder
code
was
interacting
with
ganesha,
it
was
a
different,
so
we're
trying
to
move
using
the
watch
url
approach,
so
the
the
configuration
generated
by
cfdm
could
be
good.
C
I'm
not
sure
if
there
is
any
parameter
that
we
want
to
change,
but
it's
definitely
a
good
starting
point.
D
A
comment
on
why
I
was
asking
about
the
grants
is:
is
that
in
our
scenario,
we
supposedly
don't
have
admin
rights
for
the
self
cluster,
because
in
the
o
in
in
the
you
know,
there's
the
storage
team
dealing
with
the
safe
cluster
and
there
is
the
whatever
infrastructure
or
their
team
dealing
with
openstack.
So
they
don't
have
the
administering.
D
B
B
C
I
have
an
additional
question
actually
because,
basically
for
manila
moving
to
the
cephelidium,
this
nfs
managed
by
cfadm
means
that
we
need
an
ingress
on
top
of
the
nfs
demon,
because
at
the
moment,
our
the
architecture
on
the
osb
side,
that
we
have
the
ganesha
demon
and
then
the
virtual
ip
horned
by
peacemaker
so
peacemaker.
Also
as
a
constraint
that
goes
from
the
virtual
ap
to
the
nfs
system.
The
unit
provided
by
sephonsible.
C
But
if
cepheidm
deploys
the
nfs
demons,
then
peacemaker
could
not
manage
the
the
nfs
demon
because
they
both
try
to
interact
with
the
systemd
unit.
And
it's
it's
not
the
right
way
to
do
so.
I
think
at
this
point,
the
ingress
demon
that
provides
the
beep
virtual
ipe
field
is
is
required,
and
my
question
is:
if
the
status
of
the
ingress
is
a.
G
B
F
So
francesco,
I
think
we
can
work
parallel
on
that,
in
other
words,
just
just
know
that
the
failure
numbers
develop,
throwing
away
pacemaker
and
realizing
it's
going
to
be
not
as
fast
at
first.
A
F
Be
some
long-range
challenges
you
know,
even
if
we
had
kubernetes
doing
it
instead
of
the
orchestrator.
I
mean
it's
a
it's
a
it's
a
pulling
loop,
you
know
of
some
sort,
as
I
understand
it,
you
know
so
we'll
have
to
see
how
it
works
for
aha
and
understand
it.
A
F
Yeah,
okay,
sad
ali
and
storage,
sig
kubernetes,
says
kubernetes
provides
eventual
restart,
not
h
a
so.
I
mean
that
I'm
not
taking
sides
on
this,
but
I
mean
this
is
the
challenge.
This
is
not
kubernetes,
but
it's
a
similar
approach
to
restartability
and
it's
got
to
get
good
enough
to
move
to
for
production,
but
I
I
think
I
think
that's
the
plan
so.
B
B
A
F
I
had
a
dumb
question
about
the
orchestrator
interface
and
exports
on
the
one
hand,
and
this
washer
mechanism
on
the
other.
Today
we
use
this
dbus
command
to
update
exports
today
ancient
history
into
the
present
and
it's
synchronous
so
that
we
we
find
out
whether
the
export
was
successful
or
not.
At
the
same
time
that
we
issued
the
deepest
command
manila
share,
container
runs
debuss.
F
It
we
see
right
back,
I
mean
it
sends
a
signal
over
to
the
canasha
container
through
the
host
and
using
ssh
chips
are
more
than
one
host
involved
and
immediately
there's
a
return
code
that
says
either
the
export
took,
or
there
was
some
error
and
we
see
what
the
error
is
back
on
manila
share.
If
it
fails,
is
it
correct
that
with
the
orchestrator
export
command,
we
have
an
asynchronous
term
facebook
and
we
need
to
check
on
status
afterwards
using
watch,
url
or
some
polling
mechanism,
or
something
or
another.
F
Yeah,
so
we
can't
just
substitute
the
stuff
out
step
orchestrator
command,
where
we
have
the
best
command.
Today,
and
no
I
mean
we're
going
to
have
a
little
more
complication
there
of
some
sort
and
just
getting
on
what
the
what
the
challenge
is
there
not
overwhelming,
probably
and.
B
F
A
You
you
could
call
f,
arts
restart,
nfs,
dot,
nfs
name,
but
that's
also
asymptotes,
because
we
can't
synchronously
restart
all
the
demons
on
all
hosts
because
of
timer
dishes.
We
would
block
the
manager
command
handling
thread
for
for
the
duration
of
the
ssh
timeout
for
all
the
hosts
and
some
and
that's
just
too
long.
F
Okay-
and
that
reminds
me
as
well
for
exporters,
we,
you
know
being
able
to
update
change
log
levels
or
whatever
without
actually
restarting.
The
daemon
would
be
nice
to
have
being
able
to
update
exports
without
restarting
the
daemon
and
breaking
the
daily
path
is
essential,
and
we
have
that
with
debuss.
F
Today,
rona
raja
told
me
that
var
show
was
working
on
that,
I'm
not
sure
where
I
I
heard
that
she's
back
in
school
or
something
I'm
not
sure
where,
where
we
are
on
that,
but
that
that's
a
an
essential
condition
is
that
we'd
be
able
to
update
exports
without
breaking
the
data
path
right.
So
presumably
we
can't,
if
we
do
the
equivalent
of
a
sig
hup
to
reload
config,
we
don't
want
to
actually
restart
the
process.
B
Yeah
and
there's
some
subtlety
there
that
I
don't
quite
remember
if
I
I
think
it's,
if
you
add
an
export,
you
just
poke
the
notify,
object
and
it'll
add
the
export.
But
if
you
modify
or
remove
one,
then
there's
a
reload
or
restart.
But
I
forget
exactly
there's
some
there's
somewhere
in
this
there
yeah
so.
F
B
F
C
Which,
in
theory,
should
should
work
using
the
watch
url
approach,
because
it's
a
ganesha
itself
that
use
this
approach
to
reload
using
a
seed
up
the
the
config?
F
Okay
thanks,
I
actually
have
to
run
off
to
an
appointment,
but
for
letting
me
drop
in
here,
I'm
in
good
hands
with
all
these
good,
most
peopside
people,
just
cool.
A
A
Thank
you
very
much.
Oh
wait.
We
have
the.
We
have
the
link
already
perfect.
Where
does
your
watch
perfect?
Okay,
anything
else
regarding
nfs.