►
From YouTube: Ceph Orchestrator Meeting 2022-03-29
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/contribute/
What is Ceph: https://ceph.io/en/discover/
A
All
right,
I'll
probably
start,
I
don't
have
any
topics
in
the
other
pad
for
this
week.
There's
nothing
new
that
came
up.
We
can
maybe
talk
a
little
bit
about
the.
A
The
dependencies
thing
that
rido
is
dealing
with
right
now.
A
B
This
is
this
is
basically
the
pull
request
right
now
we
are
having
well.
We
have
an
issue
open
with
vstart,
which
consists
basically
that
if
you
try
run
or
to
create
clusterism,
restart
and
enabling
self
adm,
this
fails-
and
this
happens
because
this
start
is
not
able
to
load
dependencies
of
the
selfie
dm,
and
this
is
closed
mainly
because
these
dependencies
are
not
installed,
since
they
are
run
time
dependencies.
B
B
A
All
right,
yes,
that's
kind
of
where
we're
at
with
that.
I
posted
the
two
sort
of
relevant
things
to
look
at
in
the
other
pad,
so
there's
a
one
floor
request
that
meant
to
add
the
dependencies
and
then
there's
the
second
one
is
a
issue
comment
or
not,
but
it's
comment
on
a
passport
request
and
that
sort
of
stuff
essentially
saying
that
they're
having
issues
getting
vstar
to
work.
Now
that
that's
been
added.
B
A
Yeah
so
there's
two
sort
of
things
going
on
that
we
have
to
figure
out,
so
one
is
in
the
short
term.
We
have
to
get
these
start
working
again.
I
and
also
people
use
that
development
tool
so
either.
If
we
can
maybe
get
a
quick
solution,
that's
like
okay
or
if
we
have
to
revert
that
sword
for
the
time
being,
just
to
get
that
sort
of
fixed
and
then
maybe
we
can
come
something
better
and
then
second
part
is.
A
We
have
to
actually
have
a
long-term
solution
for
where
we
want
to
put
runtime
dependencies.
It
seems
like
we
have
places
for
build
dependencies,
but
there's
not
really
anywhere
for
a
runtime
dependency
like
this.
So
in
future
cases
we
have
to
figure
that
out
as
well.
B
B
So
it's
not
that
simple,
the
the
the
best
thing
would
be
like
if
we
can
use
the
same
safe
spec
that
in
in
order
to
install
this
runtime
dependencies,
if
you
would
like
to
like,
add
some
flag
to
the
install
depths
script.
So
you
specify
like
runtime
something
like
this
and
then
you
get
all
the
runtime
dependencies.
Also,
if
you
want
to
run
the
v-start.
C
But
they
shouldn't
be
on
that
yeah.
I
just
read
his
comment.
It
seems
correct
to
me
I'm
trying
to
catch
up
on
this,
so
I'm
just
trying
to
ex
read
skim
this
spec
file
quickly.
C
There
do
seem
to
be
some
runtime
dependencies
already
there
in
the
various
sub
packages
to
both
of
these
just
apply
to
the
manager.
Only.
C
I
I'll
also
admit
to
have
having
zero
understanding
of
how
this
interacts
with
vstart.
So
do
correct
me
if
I'm
saying
stuff
that
doesn't
seem
so,
but
just
from
an
rpm
spec
file,
point
of
view.
C
You
know
it's,
you
know
a
runtime
dependency
should
be
where
it
belongs
in
a
requires
line
for
the
thing
that
requires
it.
B
Or
ceph
adm,
if
you
go
to
the
spec
to
this
manager,
staff
adm
section
we
do
have
this
requires
lines.
B
But
to
be
honest,
I
don't
know:
when
are
these
dependencies
installed
if
it
is
done
during
the
build
or
or
they
are
just
installed,
when
you
are
building
some
image
or
downloading
in
the
package,
so
I
don't
know
how
this
dependencies
are
handled.
A
B
A
C
C
And
here
it
is
requires
async
ssh.
Oh
so
you
already.
C
B
Them
yeah
the
dependency,
but
these
are
runtime
dependencies.
So
right,
I
imagine
that
when
you
install
the
deviant
package
or
the
rpm
package,
the
our
vm
will
get
this
dependencies
and
install
them.
But
if
you
build
the
cef
from
sources
and
then
you
write,
you
run
the
vstart.
B
C
B
So
the
the
the
the
workaround
I
provided
in
this
request
should
work,
because
if
you
are
building
ceph,
you
will
get
these
dependencies,
but
strictly
speaking
they
are
not
build
dependencies.
So
kefu
chai
is
is
right
because
you
know
they
are
not
build
dependencies,
but
at
the
same
time
we
don't
have
right
now
any
mechanism
to
put
somewhere
these
dependencies.
So
anybody
who
wants
to
run
vstart
will
get
them
automatically,
or
at
least
some
just
tell
them.
C
B
B
Is
I
mean
yeah,
this
sorry
adam?
I
mean
this
kind
of
issues
you
are
having
with
cfdm
could
happen
happen
with
any
python
model,
because
the
other
binaries,
since
they
are
cnc,
plus
plus
binaries.
If
you
don't
have
some
dependency,
you
will
not
be
able
to
link
the
binary,
but
python
models
are
dynamic
and
so
far,
if
you
don't
call
the
import
or
whatever
it
will
not
fail.
A
Maybe
an
alternative
could
also
be
for
the
time
being,
is
I
know
for
some
other
imports.
We
like
handle
exceptions.
If
they
don't
work,
we
could
like
try
to
import
it
and
then,
if
it
doesn't
work,
just
I
mean
we
just
pass,
it
don't
do
anything
any
exception
and
then,
in
the
part
where
we
use
it,
we
could
also
handle
an
exception
there,
and
then
we
could
work
around
it.
A
C
A
Right
now,
because
it's
been
broken
for
so
long,
but
if
we
break
restart
in
general
by
breaking
like
what
late
network
does
then
that's
causing.
D
A
If
you
can
reproduce
the
issue,
then
you
could
test
something
like
that
again,
just
like
accept
their
handling
import
errors
for
not
sort
and
then
anywhere
we
use
not
sort
or
maybe
you'll
set
a
variable
or
something.
B
A
A
Probably
better
actually
than
just
exception,
handling
everywhere.
It's
just.
It
feels
the
first
time
importing
just
have
some
sort
of
mock-up
thing
that
we'll
call
instead
yeah.
I
think
it's
yeah.
A
I
think
we
do
that
for
something
else.
I
don't
remember
what
thing
it
was,
but
I'm
pretty
sure
I
think
it
might
have
something
related
to
a
synthesis
h
that
we
had
to
do
that
for
because
only
in
like
pie,
three
seven
onwards
or
six
onwards
or
something
I
don't
remember
anymore.
I
think
we've
done
that
before.
A
That
could
be
our
good
short-term
solution,
get
it
working,
so
people
aren't
are
unblocked
and
then,
in
the
long
term
we
can
try
to
coordinate
with
people
who
actually
understand
vstar,
maybe
figure
out
where
to
put
runtime
dependencies
for
the
python
modules.
B
E
A
All
right
you
handle
that
stuff.
A
D
Yeah
I
had
a
question
I
mean
I
used
the
sephardium
flag.
I
mean
I
just
manually
install
these
missing
packages
with
restart,
but
what
do
you
guys
typically
use
for
developing.
A
I
personally
have
been
using
kcl.
I
just
started
vms,
like
vlog
vms
and
doing
it
that
way,
I
think
that's
sort
of
common
or
I
know
I
think,
don't
you
do
the.
E
A
E
A
Yeah
again,
that's
part
of
the
issues
nobody
in
the
team
uses.
These
starts
too
much,
so
we
don't
notice.
I
guess,
if
we're
installing
things
that
won't
work
by
default
and
pstart
and
it
causes
issues
for
people.
C
Yeah
one
thing
we
could
consider
doing
just
further
down
the
rotor
as
part
of
this
effort
is
we
don't
have
like
requires.txt
files
for
the
python
stuff,
a
lot
of
it's
embedded
in
the
talks
files
for
tests.
We
could
consider
doing
something
like
a
requires.text
that
could
then
be
sourced
by
the
talks,
ini
and
then
other
tools
like
if
people
had
vms
and
they
were
or
whatever
and
like
pip
installing
that
stuff,
rather
than
having
it
come
through
our
pants.
Something
to
consider
as
we
look
at
this.
B
C
Okay
yeah,
so
we
do
have
some.
I
guess
it's
not
missing
like
I
thought
it
was,
but
maybe
it
could
be
expanded.
F
Yeah,
so
those
are
mostly
for
pi
test,
but
we
could
do
something
similar
to
what
stuff
volume
does,
which
is
create
a
python
virtual
environment
in
the
build
directory
using
this
requirements,
text
file
might
be
another
way
to
go
about
this.
B
You
mean
when,
when
running
restart,
creates
virtual
environments
or.
B
F
F
And
that
might
have
implications
if
we
want
to
say,
modify
the
source
code
and
then
just
fail.
The
manager
when
you're
running
in
a
vsar
environment,
I'm
not
fairly
certain
there
yeah.
B
A
All
right,
I
think,
we're
gonna
have
to
revisit
this
topic
after
getting
some
more
info
from
people
who
understand
the
building
stuff
better.
At
least
we
have
somebody.
I.
E
A
Have
probably
written
those
down
actually
that
one
at
least
the
last
one
I'll
put
in
the
other
pad,
and
you
guys
have
any
other
ideas.
Maybe
I
don't
that
I've.
B
B
A
And
then
we
get
the
mock
or
whatever
of
that
sort
as
well,
and
then
that
that
should
solve
that.
That
at
least
give
us
time
to
work
on
like
a
real
solution.
Yeah.
A
E
A
Maybe
the
only
update
is,
I
guess
we
have
the
pacific
version
of
the
keep
alive
for
ssh
requests,
sh
connections,
so
you
don't
have
to
worry
about
backboarding
async
ssh.
I
know
we
were
considering
that
as
an
option.
If
you
had
to-
and
I
know
mike-
was
testing
something
with
removing
things
from
some
some
conflict
thing
with
nfs.
I
can't
remember
anymore
and
that
didn't
quite
work.
F
Yeah
I
was
hoping
to
speed
things
up
by
removing
the
failed
daemon
from
the
ganesha
grace
db,
which
didn't
appear
to
work
out,
so
the
clients
were
still,
I
mean
the
clients
were
mounted
in
hard
mode,
so
they
still
hung,
but
it
did
raise
a
question
in
my
mind:
can
we
scale
down
the
service
even
regardless
of
aj,
so
we
had
to
explore
that
a
little
bit
yeah.
I'm.
A
C
F
F
Is
if
it
handles
it
or
not,
I
think
the
fencing
is
okay,
but
there's
something
and
it
does
remove
from
the
grace
tv
there's
just
something
that
goes
on
in
ganesha
core,
where
the
clients
just
don't
resume
reading
and
writing.
So
I'm
not
sure
the
cause
there.
You
can
look
at
it
more.
A
All
right
and
as
for
what
we're
going
to
actually
deliver
for
pacific,
is
our
plan
still
just
to
have
sort
of
the
the
host
watcher.
The
sh
keep
the
live
connections
and
just
rescheduling.
I
like
our
best
sort
of
thing
to
do
right
now.
F
A
Yeah,
I
don't
want
to
put
that
either
and
even
then
that
would
mostly
be
for
just
making
sure
you
don't
have
to
worry
about
the
refresh
times
it
wouldn't
solve
the
sort
of
the
core
issue
of
it.
It
would
just
kind
of,
I
guess,
improve
the
worst
case
time
for
it
happening
right
now.
A
We
are
risking
the
effect
you
might
have
to
refresh
all
your
demons
or
like
try
to
refresh
all
the
devices
as
well
or
something-
and
you
could
take
you
like
quite
a
while
to
have
it-
reschedule,
get
a
fest
even
but
so
we've
solved
that.
But
this
will
be
nice
if
you
could
have
some
way
to
like
I
don't
know,
do
it
a
bit
faster
still,
but
I
don't
think
it's
gonna
happen.
A
All
right
yeah,
I
just
wanted
to
pin
it
up
again.
Some
people
who
jump
here
are
interested
in
that,
let's
see
update,
there
looks
like
we're
sort
of
set
we're
gonna
do
specific
on
that.
I
think
all
well.
It's
up
to
you
to
do
is
backboard
a
couple
things
there's
any
of
that
stuff
not
merged
yet
up
in
master.
A
I
think
no
there's
a
the
host
protection
thread
is
still
not
merged.
I
need
to
do
another
qa
run,
but
I
think
it's
sort
of
ease
up
a
little
bit.
I
know
for
a
while.
You
couldn't
get
anything
through
very
really.
It
was
super
slow.
A
D
A
A
D
Okay,
and
should
I
I
mean
I
can
go
ahead
and
create
it,
do
you
want
me
to
create
it
under
self
manager
or
under
sephardium?
D
A
Yeah
I
mean
I
don't
see
right
omegas,
yes,
it
wouldn't
depend
on
any
async,
ssh
or
agent
or
anything
that
would
be
its
own
thing,
so
I
guess
it
could
be.
I
wouldn't
get
it
as
soon
as
the
other
things
like.
I
don't
know
when
this
would
get
done,
I'm
not
sure
how
much
work
it
would
take
to
get
that
working
but
yeah.
I
don't
see
why
not
it's
not
something
that
just
built
on
top
of
some
larger
things
and
only
investor.
F
A
I
know
they
that's
true
yeah.
You
know.
F
At
how
it
works,
there
is
some
logic
where
it
kind
of
starts
at
the
default
nfs
port,
and
then
it
just
kind
of
increments
forward.
I
think
the
problem
is:
is
that
we're
not
really
aware
of
which
ports
are
already
in
use
on
the
host
yeah.
A
A
F
A
I
guess
also
the
other
thing
that
makes
it
a
bit
weird.
Is
that
what
that
is
for
rgw?
You
specifically
sell
you
like.
We
want
two
of
them
on
the
same
house
like
that's,
very
explicit.
This
would
be
like
sort
of
us
deciding
that
on
our
own
yeah,
so.
A
D
I'll
just
go
ahead
and
create
a
tracker
ticket
for
now
nice.
We
want
to
track
it
for
open
stack
as
well.
The
other
question
I
had
is
we
weren't
even
able
to
mount
the
nfs
export
when
we
had
an
increased
daemon
running,
so
I'm
wondering
how
we
were
able
to
test
it.
D
A
D
F
D
And
how
do
you
create
the
cluster?
Do
you
use
the
nfs
cluster,
create
command,
or
do
you
use
the
service
specs
I've.
F
F
A
Yeah
and
that
one
in
particular
at
least,
is
in
yuri's
testing
thing,
but
the
other
one,
the
one
that
the
watcher
that
gets
it
really
20
seconds
that
one
is
still
not
merged
in
master
yet,
and
then
the
reaction
rescheduling
needs
to
be
backported
to
pacific
to
do
as
well.
So
I
get
a
test
running
today
for
this
ones.
D
A
Okay,
yeah
we'll
have
to
see-
because
maybe
this
is
something
that
only
happens
in
pacific.
F
I
think
I've
seen
similar
things,
but
it's
usually
like
there's
something
wrong
with
the
network
bridge,
specifically
the
vet,
some
resolution
problem,
or
very
occasionally,
sometimes
as
soon
as
I
configure
and
export,
and
the
suck-up
happens
for
the
daemon
that
we're
supposed
to
reload
the
export
config.
Sometimes
that
doesn't
occur,
that's
really
slow.
But
this
looks
a
little
different
because
I
think
they
mounted
the
daemon
directly
and
that
succeeded,
but
they
were
unable
to
mount
through
the
vip.
So.
D
F
D
D
Because
I
think
you
can
also
set
up
the
port,
you
want
thirteen
to
listen
on
that
can
be
configured
but
yeah.
I
I,
I
don't
think
it's
support
issue.
F
F
The
the
vip
address
the
159
address-
okay,
oh
wait,
except
for
it
did
connect
to
the
reason
by
server,
no
such
file
or
directory,
because
that
did
resolve
it.
I
mean
the
other
thing
you
could.
Try
is
just
mount
at
the
root
of
the
nfs
volume
and
see.
If
you
see
the
exports
there,
you
can
just
list
all
the
exports.
There's
like
a
pseudo
file
system
or
nfs.
D
F
D
A
A
F
D
Yeah,
I
think
we
want
to
not
just
use
the
cluster
create
because
it's
not
flexible
so.
D
So
you
want
to
like
you
said
there
might
be
something
missing,
that
cluster
create
does
and
and
individually
deploying
the
nfs
cluster
and
then
the
english
service.
On
top
of
it,
there's
certain
setups
set
up
steps
that
are
missing.
That
makes
this
not
work,
but.
F
D
A
All
right,
ronnie
you've
worked
with
the
an
nfs
module
before
right.
Yeah
changes
there
now.
Can
you
figure
out
if
there
is
anything
like
that
and
cluster
creates
just
figure
out?
If
that's
where
your
problem
is.
A
We
have
a
few
options,
I
mean
it
could
be
that
it's
that
is
working
for
mike.
So
it's
not.
It
can't
be
like
a
totally
generic
thing,
so
it's
either
that
something
that's
only
in
pacific.
So
I
assume
your
testing
is
in
pacific
or
it's
like
testing
on
master,
it's
something
that's
specific
to
the
cluster
creates,
or
it's
some
sort
of
environmental
thing.
I
like
the
networks
or
setup
or
something.
D
Yeah
yeah,
I
mean
I
tested
this
with
master
and
used
cluster
create,
and
I
had
a
similar
issue.
So
I'm
guessing
some
something
related
to
the
networking
yeah
that
mike
was
talking
about.
D
A
Yeah,
maybe
you
guys
can
coordinate
an
irc
or
something
yeah
information
maybe
could
be
useful.
E
A
No,
in
that
case,
I
think
we
can
end,
or
today.
B
I
started
working
on
the
issue.
Then
I
saw
that
there
were
like
some
attempts
in
the
past
that
were
rejected
because
in
fact,
in
sephidim,
normally
we
handled
the
output.
F
Yeah,
I
think
so
so
like
most
of
bins,
fadm,
sometimes
it's
just
freeform
human
readable
and
sometimes
it's
json
depending
on
the
command.
The
manager
expects
structured
output
on
certain
commands
and
other
commands
are
tended
to
be
user
facing.
But
we
don't
really
have
a
formatter
which
would
probably
be
nice
yeah,
because
I
know
there's
been
issues
like
specifically
on
bootstrap.
F
B
So
yeah,
and
so
maybe
this
is
like
a
different
feature-
would
be
I
in
in
this
request.
I
just
tried
to
fix
the
the
issue
with
the
messages
redirection
so
basically
to
send
info
and
the
back
messages
to
the
standard
output
and
to
send
the
errors
to
the
standard
error
which
normal
linux,
linux
or
unix
user
would
expect.
A
Comments,
I
like
briefly
remember
the
idea
of
somebody
floating
like
printing
out
the
booster
public
and
jason
in
certain
cases
if
they
had
a
flag
or
something
like
in
the
past
and
they
got
that
didn't
get
merged
in.
I
don't
remember
what
happened
with
it.
F
A
Okay,
anything
else
talk
about
foreign.
A
A
All
right
that
case,
and
here
and
I'll
see
you
all
next
week.