►
From YouTube: Ceph Orchestrator Meeting 2022-08-16
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
C
B
B
Can
you
see
the
part
yeah,
okay,
Yeah?
So
basically
as
part
of
a
five
to
three
correct
me?
If
wrong
Adam,
please,
we
need
to
add
support
for
a
rgw
multi-site
configuration
and,
if
ADM,
right
now
in
order
to
configure
rgw,
you
have
to
perform
several
steps
manually
using
using
the
rgw
admin
binary
and
it
is
little
bit
cumbersome.
So
the
idea
is
to
put
all
this
configuration
in
safe
ADM
to
make
it
easier
for
end
user
to
configure
rtw
multi-site.
B
So
there
was
a
Tracker
opened
by
Sebastian
one
year
ago,
and
here
he
put
some
some
ideas
about
how
this
could
be
implemented
and
the
main
idea
introducing
this
track,
basically,
so
that
the
new
new
layer
on
top
of
service
type
to
distinguish
between
aspects.
So
right
now
we
have
a
service
spec
generics
of
respects
to
cover
different
services
that
we
support
on
on
Seth
and
another
class.
B
That
represents
the
host
spec,
so
Sebastian
ideas
to
add
this
new
top
player,
with
the
keyword
kind,
where
you
specify
what's
the
type
or
the
kind
of
of
this
spec.
So
in
this
case,
instead
of
using
service
type,
for
example,
to
specify
the
host,
which
will
be
much
clearer
to
use
client
hosts-
and
here
he
puts
an
example
of
this
spec
that
we
discussed
later
yeah.
B
So
this
is
from
one
side
on
the
other
side,
I
respect
little
bit
the
source
code
to
see
what
we
have
right
now.
So
in
the
python
common
deployment,
we
have
a
different
classes
for
the
specs
that
we
support
as
before
the
spec,
which
is
the
largest
separate
file
host
placement,
spec
server
spec,
which
is
the
base
class
for
all
the
different
specs
that
we
support
all
of
this
and
then
in
the
interface
that
Pi
in
the
orchestrator
model.
B
B
So
this
is
more
or
less
the
current
situation.
B
My
idea
is
I
mean
the
idea
is
similar
to
what
the
Sebastian
proposed
at
the
beginning,
but
you
have
summed
up
about
how
can
we
implement
it?
So
my
ideas
to
added
a
generic
spec,
a
for
example,
7dm,
spec
or
general-
expect
I'm
not
but
I
mean
I,
don't
have
a
clear.
B
Preference
we
can
use
generic
spec
or
cpdm
spec
I
prefer
the
second
one
that's
little
bit
more
than
the
first
one,
because
it
relates
to
save
EDM
clearly,
and
this
will
be
like
the
base
class
for
all
the
different
specs
that
we
were
supporting
safe
EDM,
including
service,
pick
host
pick
and
all
the
different
new
rgw
objects
and
items
that
will
need
so
we'll
have
this
keyword,
client
or
type
or
whatever
wants
to
call
it.
B
B
Idea
was
to
include
like
separates
items
like
this
one,
for
example,
to
create
the
the
the
the
realm
I
would
go
to
track
my
directly,
so
this
one
will
be
to
create
the
real.
This
one
will
be
great
to
the
zone
and
this
one
for
for
the
service,
so
I
think
his
idea
was
to
create
like
each
element
separately
and
then,
when
you
arrive
to
the
service,
you
just
check
that
you
already
have
jig.
Somehow
you
already
have
this
real,
and
this
Zone
and
in
case
of
Zone
group
Zone
group,
created.
B
I'm
not
sure
if
this
is
good
or
bad,
that
the
alternative
would
be
like
putting
the
information,
all
the
information
related
to
the
rgw
in
in
this
entry
and
then
use
the
information
about
three
arm:
The
Zone
and
the
Zone
group
to
create
the
different
stuff.
That's
that
we
need
first
Alternatives
seems
more
flexible,
as
you
can
put
whatever
you
want
for
these
entries
and
in
the
future.
Maybe
you
we
can
extend
it
by
more
fields.
In
case
we
need
to
the
other
alternative
like
putting
everything
under
this
entry.
B
B
Pros
and
cons
I
see
for
for
both
of
them,
then
other
things
that
we
have
to
keep
in
mind
like
the
the
RCW
demonstration,
must
be
done
in
a
specific
order.
So
if
we
go
with
the
alternative
to
have
separates
entries,
then
we
we
have
to
ensure
that
we
are
creating
things
in
the
in
the
right
order.
So
first
we
have
to
create
the
realm.
Then
the
Zone
group,
then
the
zone
you
can
not
just
go
into
process
of
the
the
ammo
spec
one
by
one.
B
B
Think
right
now
we
are
using
default
preamp
default
rgw
for
everything
that
shouldn't
break,
but
we
have
to
make
sure
that
we
are
not
breaking
or
old
functionality,
I'm,
not
missing
any
existing
HW
installation,
for
example,
case
of
upgrades
and
in
case
of
the
users
already
have
like
three
uncreated,
which
shouldn't
mess
with
what
the
user
already
have.
B
A
Yeah,
okay,
so
what
our
initial
thoughts
is?
The
the
most
important
thing
that
I'm
I
was
worried
about
is
the
backwards
compatibility
like
I
know
when
I
was
looking
at,
say,
tracker
with
the
service
types
and
stuff
like
even
like
the
most
like
the
host
specs
are
even
a
little
bit
different
in
this
proposal.
A
However,
even
if
like
it
wasn't
the
problems,
if
we
could
like
migrate,
the
effect
that
already
exists,
it
just
sort
of
I,
don't
know
Annoying
I
guess
for
the
users.
If
we
just
change
the
way
you
have
to
format
like
the
specs
like
that,
because
it's
been
a
few
years
now,
I
think
they
looked
sort
of
like
this
yeah.
A
D
D
If
you
don't
mind
me,
I'm
gonna
talk
for
a
moment,
so
two
big
things
jump
out
at
me.
One
is
I
generally
like
the
idea,
but
of
course
this
is,
this
is
making
Seth
Adam
even
more
like
kubernetes,
especially
with
the
use
of
the
word
kind.
B
These
points
before
we
continue
I
think
Sebastian
selected,
the
the
word
kind,
because
can
you
see
this
is
this?
Is
the
The
Rook
spec
and
they
are
using
kind
for
the
same
or
similar
thing
I
think
he
borrowed
the
world
like
for
anybody
familiar
with
Rook
to
be
able
to.
D
D
Gvk
stands
for
group
version
kind,
so
an
API
version.
The
thing
before
the
slash,
so
in
this
case
sef.broke.o,
is
the
group.
The
thing
after
the
slash
is
the
version
in
kubernetes,
they're,
typically
stuff,
like
V1
or
V1,
Alpha
One,
that's
not
so
irrelevant.
The
point
is
together
the
three
make
like
a
three
Tuple
to
uniquely
identify
a
resource
so
because
resources
usually
come
as
a
group.
D
That's
why
it's
API
version
is
one
thing
and
then
kind
kind
of
distinguishes
the
individual
resource
types
that
come
through
the
group
inversion,
which
is
all
to
say
that
it's
a
very
important
for
kubernetes,
because
kubernetes
allows
you
these
days
to
Define
your
own
resources.
The
Rook
is
defining
resources.
Other
operators
and
controllers
are
defining
their
own
resources.
D
So,
if
ADM
doesn't
need
to
be
that
generic,
but
there's
the
reason
they
did,
this
is
because
it
allowed
them
to
make
backwards
and
compatible
changes.
So
if,
at
one
day
they
decided
there
would
no
longer
be
you
know,
pull
then
you
would
say
change
the
API
version
to
V2
and
you
would,
you
know,
no
longer
have
to
support
the
pull
field.
D
That's
just
an
example.
So
does
that
all
make
sense?
D
Yes,
yes,
yes,
no!
Okay,
so
I
don't
know
how
far
Sebastian
was
thinking,
whether
he
thought
that
there
would
be
a
need
to
support
cube-like,
Behavior
I,
don't
know
I'd
say
the
the
interesting
part
is
versioning
and
the
kind
stuff.
So
one
other
thing
is
depending
on
how
much
you
want
to
make.
This
like
kubernetes
is
like
if
you
want
to
use
the
word,
there's
good
things
to
using
the
word
kind,
which
is
like
familiarity
if
you've
used
play
rookie
like
oh
kind,
it's
kind
of
equivalent
to
what
kubernetes
does.
D
On
the
other
hand,
it's
not
going
to
be
exactly
like
kubernetes,
so
there's
a
risk
where,
if
you
use
the
word
kind,
it'll
confuse
people,
because
it's
not
going
to
be
exactly
like
kubernet
I,
just
want
to
throw
that
out
there.
If
you
ever
have
any
questions
about,
do
Behavior,
don't
hesitate
to
ping
me.
B
Okay,
thanks
John
yeah
I,
see
I,
mean
I,
made
the
comments
because
I
think
in
this
track,
or
not
sure.
If,
on
this
tracker
or
another
one
Sebastian
mentioned
that
Rook
team
is
has
already
working
on
something
similar
and
he
puts
right
to
this
entry.
You.
D
B
D
D
Thing
you
said
that
I
want
to
point
out
is
the
order
that
would
be
now
I,
don't
know
what
Brooke
is
doing
if
they're,
if
it,
if
the
order
matters
to
Rook,
then
they
have
a
bug,
because
the
kubernetes
declarative
approach
is
a
very
much
like
it
shouldn't
matter.
What
object,
what
order
the
crds
were
created
in,
because
the
reconcile
Loop
should
take
care
of
that,
for
you.
D
C
B
That
that
was
the
points
basically
from
my
comments.
I
mean
like
here,
for
example,
if
the
user
introduced
it,
this
spec
and,
for
example,
puts
the
Zone
before
the
real
internally
in
safety
DM.
We
shouldn't
error
because
right
the
the
realm
goes
after
the
Zone
should
be
more
clever
and
process,
first
real
and
the
zone
to
be
more
user
friendly
because
spec
and
it's
correct,
but
that
will
require
more
processing
on
the
safe
ATMs
site.
B
Obviously
right
so
here
you
have
something
so
I
think
Sebastian
just
look
at
this
entries
and
he
copied
something
similar
on
surf
EDM,
because
here,
as
you
can
see,
this
is
the
Zone
group
and
then
in
the
spec,
a
link
to
the
tutorial
to
the
correspond
Realm
yeah
in
this
Zone
at
some
point
here
is
the
zone.
Then
he
must
The
Zone
group
yeah.
He
linked
to
the
Zone
group
at
this
point,
and
some
group
is
linked
to
the
real.
A
I
was
sort
of
imagining
us
handling.
These
things
is
that
the
kind
of
way
we
have
that
apply.
Services
we'd
also
have
like
a
some
other
apply
thing
that
runs
before
that,
and
then
it
would
go
through
whatever
respects
we
have
for
these
types
of
resources
like
if
we
have
a
Zone
group
spec
and
a
Zone
spec
I
would
always
do
the
round
ones.
First,
then
they
would
do
the
Zone
ones
and
the
Zone
groups
and
then
make
sure
that
all
the
stuff's
always
done
in
that
order
and
it'll
be
done
before
the
services.
A
So
we'd
never
have
to
worry
about
anything
stored
and
basically
I
would
just
try
it
and
then,
if
it
fails,
it
would
say,
raise
a
health
warning
or
something
saying
like
I
couldn't
make
this
one.
If
it
succeeds,
I
I
mean
I
thought
it
started.
A
My
head
was
that
it
would
just
sort
of
get
rid
of
the
spec
at
that
point,
because
this
should
only
be
a
one-time
thing,
so
we
just
make
the
thing
and
then
we're
sort
of
done
with
that
spec
get
rid
of
it
and
if
not,
if
it
fails,
then
we
keep
the
spec.
But
we
raise
the
health
warning
to
say
like
we
want
to
make
this
thing
for
you,
but
we
can't.
C
A
I
mean
we
could
just
they've
I
mean
like
so
this
is
like
one
big
spec.
We
could
just
save
them
all
as
a
bunch
of
separate
specs,
and
then
we
just
eventually
like
the
way
we
use
search
right
now
you
can
search
like
buy
like
a
service
type
whatever,
when
you're
looking
for
specs
we'd
have
to
have
some
way
of
searching
for
like
the
type
of
spec
racism.
The
kind
now
say
they
say
it
here.
A
I
mean
I
want
to
get
all
the
rgw
realm
specs
and
I'm
going
to
go
through
and
I'm
going
to
try
to
create
all
those
Realms,
I'm
gonna
get
all
the
RGB
Zone
ones
and
so
on,
and
that
would
be
a
separate
thing
that
happens
before
we
make
any
Services,
because
if
there
is
an
rgw
service,
that's
depending
on
these-
and
that
has
to
happen
be
made
after
these.
These
zones
and
Zone
groups
and
stuff
so
yeah
I,
don't
think
it
would
be
a
a
big
problem.
A
B
Is
something
that
we
have
we
have
to
solve
for
now,
and
we
may
have
like
a
lot
of
a
lot
of
entries
of
this
kind
in
different
order
in
different
relationship
between
the
Realms
zones,
Zone
groups,
so
we
have
first
to
make
sure
that
the
the
structure
is
correct
before
creating
anything.
Otherwise
we
may
end
up
like
with
partial
creation
like
we
have
the
real,
but
then
the
zone
or
the
Zone
groups
were
not
created
because
because
they
are
not
correct,
so
I
think
we
should
make
it
kind
of
atomic.
B
If,
if
it
succeed,
then
it
succeeds
for
for
everything.
If
it
fails,
then
we
shouldn't
have
like
to
roll
back
to
delete
whatever
a
realm
has
been
created.
D
Yeah
I
mean
it
shouldn't
deploy
Services
until
all
the
other
pieces
have
been
satisfied,
yeah
I
mean
the
other
thing.
Is
you
could
you
could
do
certain
things
lazily
right
like
if
you
create
I,
don't
know
I,
don't
know
enough
about
multi-site,
but,
like
suppose
that
you
define
you
submitted
the
yaml
for
for
Zone
here,
that's
the
only
thing
you
did
well.
If
there's
no
Services
backing
that
zone,
do
we
need
to
do
anything?
Maybe
it
doesn't
do
anything
until
it
sees
as
an
rgw
service?
That's
going
to
make
use
of
those
things.
C
A
A
B
Out
anyway,
John
I
mean
that
this
is
good
idea,
but
when
we
have
the
problem
that
in
case
you
do
it
lazily,
maybe
you
fail
very
late,
because
whatever
a
field
in
the
zone
was
not
correct
and
you
would
not
swarm
the
user
as
soon
as
possible
that
the
spec
he
was
just
trying
to
use
is
not
correct.
D
Yeah
totally
I
I
agree.
The
issue
is
I,
don't
know
enough
about
the
specifics
of
our
GW
multi-site
to
make
a
guess
at
whether
like
should
should
this
be
treated
as
an
active
thing
like
like
a
service
I
want
to
a
service
is
turned
into
running
processes.
D
C
C
B
D
Yeah,
like
I,
said
it
it.
It
kind
of
depends
on
whether
this
is
like
an
active
property
of
the
system
or
more
of
a
passive
property
of
the
system.
Whether
you
can
apply
it
lazily
or
not,
and
I
I
don't
know
I
should
what
I
should
do
is
if
we're
going
to
have
this
conversation
again
is.
We
should
have
probably
all
read
up,
at
least
on
the
docs,
the
official
docs,
and
how
to
do
it
manually.
So
then,
we
are
a
little
bit
more
aware
of
what
the
Yeah.
B
D
My
questions
are
basically
is
the
structure
that
Sebastian
proposed
makes
sense
for
rgw
in
this
ADM
context,
and
is
it
even
neces?
Is
it
even
necessary
to
implement
rgw
multi-site
I'm,
not
saying
that
we
shouldn't
do
it
I'm
just
saying
we
should
at
least
think
about
whether
these
are
orthogonal
tasks
or
Sebastian
kind
of
glued
them
together,
because
it
appealed
to
him.
B
I
think
the
need
right
now
is
basically
in
order
to
configure
rgw
multi-site.
We
have
to
issue
a
lot
of
different
commands
to
create
to
realm,
to
create
the
zones
his
own
groups
manually
and
then
use
safe,
ADM
at
at
the
last
instance
to
to
create
the
service.
So
maybe
this
kind
of
we
have
another
Administration
tool
which
is
set
free
DM.
But
yes,
it
is.
Is
it
partially?
So
it's
from
another
perspective.
It
is
just.
D
Hypothetically
and
I'm
not
saying
we
should
do
this,
but
I
want
you
know
and
understand
like
if
they
everything
could
be
put
into
today's
rgw
resource.
Maybe
it
would
be
ugly
or
clunky
or
whatever,
but
if
it's
do
a
bowl,
then
we
need
to
at
least
consider
the
fact
that
you
know
the
group
The
or
the
kind
stuff
might
be
cool.
It
might
not
be
strictly
necessary
either.
A
I
mean
it's
definitely
possible
to
throw
it
all
in
there.
You
could
just
have
a
really
good
argument.
Spec
that
has
all
the
things
listen.
We
could
just
we
can
already
list
the
ramen's
owners
up
there.
We
could
just
add
some
functionality
that
just
tries
to
create
it
for
you
first,
if
it
doesn't
exist
yet
and
right
that
could
be
all
it
does.
That's
the
possible
way
to
do
this.
It's
definitely.
B
Yeah,
as
as
I
said
at
the
beginning,
that's
another
way
to
go
with
this.
The
risk
in
that
case
is
that,
if
they
start
adding
more
functionality
in
the
future,
so
maybe
at
some
points
this
entry
will
become
huge
with
a
lot
of
configuration.
That
should
be
hard
to
maintain
right.
D
More
generic
and
flexible
in
the
future
exactly
I
I.
My
point
is:
we
need
to
decide
whether
we
want
to
make
it
a
requirement-
or
we
just
say:
oh
well,
they're
both
good
things,
but
we
can
do
our
GW
multi-site
in
the
short
term.
Without
it
and
come
back
and
do
this
later,
I
don't
know
I'm
just
I'm,
throwing
that
out
there,
because
yeah
I've
been
on
lots
of
projects
for
people
blob
stuff
together,
and
it's
because
it
sounds
cool
and
I'm.
Just
like
okay.
B
Yeah
yeah
you're
right
that
that
depends
I,
think
if
we
had
some
more
input
from
product
management
like
what
kind
of
features
they
foresee
like
in
the
next
year
and
next
two
years,
so
we
can
add
more
information
if
we'll
be
adding
more
functionality
to
this,
or
it
should
be
just
like
this.
Just
to
create
the
basic
I
mean
if
it
is
just
a
great
Basics,
then
maybe
it's
easier
just
to
put
whatever
information
we
need
in
this
rgw
service
pick
and
go
with
it.
It
will
be
much
more
easy.
A
When
you're
talking
about
sort
of
like
a
downstream
thing,
I
can
say
with
what
I've
talked
to
with
them,
because
obviously
that's
where
this
is
getting
into
the
push
from,
but
it
should
be
good,
useful
production
as
well.
Is
they
almost
just
want
to
be
able
to
sort
of
do
this
in
less
steps?
I
think
is
the
important
thing
to
them
right
now,
because
it's
just
like,
as
you
were
saying
earlier,
it's
just
it's
a
lot
of
steps
to
go.
You
know
create
all
the
things.
A
Do
all
that
I
think
it's
just
more
just
making
this
a
bit
easier
to
deploy,
because
it
is
annoying
to
the
but
I've
done
it
before
I've
deployed
multi-sites
and
it
takes
a
while
there's
a
lot
of
stuff
to
do
it.
So
this
would
just
make
it
easier.
I,
don't
think
they
care
too
much
how
we
do
it
I
think
they
just
want
it
to
work.
I
think
this
stuff,
where
we're
saying
like
oh
we're
going
to
do.
This
is
more.
What
do
we
want?
What
do
we
like
better?
In
our
thing?
That's.
B
To
think
that,
if
we
go
with
this,
let's
see
the
simple
way
just
put
everything
in
in
this
service.
If
at
some
point
later,
we
want
to
change
this
flexible
scheme,
then
we
have
to
be
backwards
compatible
with
whatever
we
introduced
in
the
first
implementation
right.
D
So
getting
back
to
my
earlier
point,
if
we
Implement
kinds,
we
should
at
least
also
Implement
verbs.
We
can
probably
ignore
groups
because
that's
not
Steph
ADM
is
not
going
to
be
kubernetes.
It's
going
to
not
going
to
be
that
specific.
But
if
we
do
versions,
then
at
least
you
can
say
oh
well,
this
is
a
kind
of
you
know:
rgw
thingy,
one
versus
Persian
and
kind
rgw,
thingy
version
too.
B
Yeah,
that
was
the
last
comments.
I
put
your
ownership
of
community.
Should
we
added
new
version
feel
to
the
spec.
I
was
thinking
in
the
same
line
as
the
argument
right
now.
If
it
makes
sense
or
not,
maybe
right
it's,
it
is
good
idea
to
have
something
there.
So
this
way
with
Detective
we
are
on
version
one
or
whatever
and
then
process
the
the
spec
accordingly.
C
A
The
way
we
could
do
in
the
meantime,
because
I
assume
this
was
confident
the
weekly
next
week,
for
that,
if
you're
still
looking
at
this
stuff
we
could
do
is
because
this
is
mostly
almost
a
front-end
discussion.
It
seems
to
be
where
all
the
talking
points
are
we
could.
What
you
maybe
could
do
is
try
to
do
sort
of
the
back
end
stuff
like
so
we
were
saying
before:
there's
gonna
be
like
some
function
before
the
apply
services
that,
like
you,
know,
does
all
this
stuff.
A
Maybe
you
can
see
like
whatever
the
simplest
possible
front
end
is
like.
Maybe
that
is
to
shove
me
everything
in
the
rgw
spec
and
just
have
some
like
sample
stuff
or
whatever,
and
then
we
can
see
like
this
is
how
it
would
work
in
the
back
end
like
what
kind
of
front
end
do
we
really
want
around
this
yeah?
Just
maybe
so
we
can
get
the
full
picture
of
what
we're
doing,
and
then
it
also
because
I
mean
the
one
thing
with
the
I
guess.
A
If
we
like
finalize
doing
it,
one
way
in
the
front
end
is
then
it
would.
You
have
to
then
worry
about
well
as
soon
as
anyone
starts
using
that
we
now
have
to
support
it
like
almost
not
forever,
but
for
a
while,
and
even
if
we
do
want
to
change
it,
so
I
kind
of
want
to
make
that
decision
almost
like
last,
like
you
know,
decide
everything
like
how
it's
going
to
work
behind
the
scenes
and
then
what
we're
going
to
actually
do
and
then,
at
the
very
end,
all
right.
A
A
Yeah
I
mean
I,
think
that's
exactly
what
I
was
sort
of
thinking
is
that
we
we'll
figure
out
all
this
stuff
and
then
we'll
do
at
the
end,
we'll
figure
out.
What's
the
best
way
to
present
this
to
people
and
everything,
so
we
can
yeah.
C
B
A
Word
of
warning
on
that
is
that
if
there
are
no
osds
in
your
cluster,
that
command
just
hangs
forever,
so
you
have
to
probably
some
sort
of
time
out
really.
Yes,
all
the
videos
Gateway
admin
commands
do
that
they
will
hang
forever
if
there's
no
osbs,
so
most
people
in
their
clusters
have
osds,
but
we
do
have
to
be
aware
of
it
that
we
have
to
have
time
outs
on
all
those
I.
Think,
okay,.
D
Yeah
I'm,
assuming
ultimately
we'll
you
know
in
the
short
term,
yeah
sure
we
can
reuse
a
an
existing
tool,
but
it
would
make
probably
sense
to
either
have
the
logic
in
apis
because
I'm
assuming
it's
not
doing
anything
magical,
it's
either
creating
rados
objects
or
whatever
just
to
configure
everything.
And
so
you
know
we,
we
probably
don't
want
to
add,
like
a
container
just
to
run
that
one
command
no.
D
I
know,
but
to
call
the
binary
I
need
to
call
it
from
something:
I,
don't
want
to
see
a
bunch
of
shell
outs
inside
of
like
a
manager,
module
or
something
it
doesn't
make
a
heck
of
a
lot
of
sense.
In
my
opinion,.
A
A
I,
just
I
just
put
the
link
in
there.
Okay,
I
baby,
it's
even
possible
that
we'd
end
up
deciding
instead
of
implementing
stuff
the
way
we're
doing
with
the
four
because
I
know
that's
the
way
I.
What
I
liked
you
before
was
something
that
Dan
did
a
long
time
ago,
where
he
had.
It
was
just
actually
doing
basically
what
drama
where
he
just
like
runs
these
little
sub
process
commands
that
go
and
run
radio
gatway
admin
commands.
C
A
It
was
very
ugly
and
that's
part
of
why
maybe
we
removed
it
yeah,
but
because
this
module
exists,
maybe
there's
a
possibility
of
getting
that
sort
of
functionality
in
there.
So
that's
like
the
module,
where
you
do
all
like
the
ugly
stuff
to
make
this
rhythmic
Gateway
how
to
make
commands
work
and
then
we
could
nicer
one.
Maybe
we
can
contribute
to
that,
but
I
have
to
see
what's
actually
I,
don't
think
know.
What's
in
there
I
haven't
did
the
argument
should
be
much
longer
yeah
right.
D
A
A
This
module,
there's
something
that
we
use
the
NFS
module
that
accuses
orchestrator
module
to
do
stuff.
We
could
maybe
have
the
tech,
video
module,
use
the
rgw
module
yeah.
B
D
A
A
A
C
A
A
C
D
A
B
A
C
A
C
A
Yeah
I
guess
we
can
start
here,
though
we
can
start
with
doing
figure
out
this
back
end
stuff
and
how
we're
going
to
do
the
the
commands
and
everything
and
then
we'll
finish
up
with
the
The
Front
End
discussion,
add
a
bit
more
time
to
think
about
it
because
I
think
that's
again.
Actually
this
oddly
enough,
the
more
important
one,
because
that's
the
one
that's
harder
to
change
once
it's
put
in.
C
B
And
then,
maybe
if,
meanwhile,
you
can
get
more
information
in
from
product
management
in
Downstream
to
see
what
what
plans
do
they
have
for
for
rgw
multiset
support
in
the
new
term,
it
will
be
adding
more
functionality
or
not
just
chopping
them
and
see
if
they
have
something
in
mind
or
basically,
the
I.
A
Think
the
most
likely
thing
that
will
end
up
playing
out
would
be
that
they
want
it
to
be
easier
to
do
and
then
once
people
start
using
it,
then
they
will
find
out.
There
are
things
missing
or
things
they
want,
but
we
won't
know
what
those
things
are
until
they
try
it
yeah
we're
almost
going
to
be
forced
into
like
a
best
effort.
First
attempt,
so
we
just
try
to
like
make
it
easy
to
set
up
a
basic
multi-site
thing,
and
then
people
will
try
to
be
like.
A
C
B
And
connecting
customers
will
start
asking
for
more
features,
more
options.
A
When
people
start
using
an
ml
that
they
want,
but
it's
hard
to
say
beforehand,
yeah
but
I
think
what
we're
doing
here,
where
we're
just
trying
to
create
the
realm
zone.
So
definitely
a
good
start
and
I'm
off
to
look
again
at
the
steps
to
make
the
ulti
slide
and
see.
If
there's
something
else,
maybe
that
we
should
be
doing
or
if
this
is
as.
B
Great
and
then
I
think
we
we
are
on
the
same
page.
We
can
start
well,
I
can
start
working
on
the
back
end
stuff.
I
will
have
more
deeper
look
to
this
model
to
see
what
it
does
and
if
we
can
use
it.
Otherwise,
I
will
just
try
to
edit
functionality
in
the
same
model,
whatever
I
need,
and
then
we
can
postpone
the
discussion
for
the
front
send
for
the
next
week.
B
C
B
A
So
I'll
just
mention
it
briefly,
so
this
was
a
Tracker
issue
that
popped
up
last
week
or
something
there's
this
guy
who
just
he
can't
plays
osds
I.
Think
it's
something
privy
to
implement
our
distribution
thing.
Maybe
I
know
he's
on
like
a
rocky
Linux,
whatever
I
don't
know
the
details
of
why
exactly
it
doesn't
work,
but
I
know
that
you
cannot
deploy
any
osds,
because
it's
getting
in
this.
You
know
it's
on
the
tracker,
but
it
is
giving
him
some
error.
A
Let
me
go
find.
This
error
could
not
find
secret
mount
in
proc,
stash
self,
C
group,
and
so
because
we
put
that
in
everywhere
automatically
if
you
have
a
podman
version
or
a
podman
above
a
certain
version,
you
just
can't
deploy
anything
on
that
host.
Obviously
he
has
a
mixture
of
different
OS
types
in
his
cluster.
For
some
reason,
but
based
on
this
one
in
particular,
you
can
is.
D
D
A
A
Yeah,
so
my
question
was
not
necessarily
about
the
distro
itself
or
why
the
issue
came
up
but
more
about
what
should
be
done
about
it,
so
I
linked
in
there.
Some
potential
pull
request
fix.
They
basically
just
copied
the
way
we
fixed,
the
diff
or
the
init
container
or
whatever
the
the
flag
was
called.
A
And
basically
all
it
does
is
it
just
gives
you
a
config
option,
and
then
we
have
some
other
option
inside
video
that'll
say
like
I
want
to
use
this
flag
and
so
then
it'll
deploy
all
the
advantage
out
the
flag.
Okay,
so
it's
very
simple
patch
it
just
and
it
is
from
the
config
option
on
and
then
it
should,
if
you
just
as
soon
as
you
redeploy
the
containers
they
should
be
deployed.
Without
this,
it
should
fix
the
problem.
A
I
guess
the
question
I
sort
of
had
is.
Is
this
the
way
we
kind
of
went
before
because
I
feel
like
if
we
keep
doing
this,
where
we
slowly
like?
Oh,
let's
add
this
flag
into
this
container,
because
it's
good
and
whatever
then
we
find
out
later
it
doesn't
work
on
certain
people.
We're
gonna
have
to
do
the
exact
same
thing
again.
We
just
had
a
new
flag
that
says
now
we're
not
going
to
tell
you
like.
A
Don't
use
this
thing
right,
and
can
we
eventually
come
to
some
generic
way
of
handling
situations
like
this?
Are
we
stuck
just
just
just
be
really
careful,
I
guess
adding
new,
automated
Flags.
A
So
this
the
well
I
forget
so
some
guy,
some
outside
contributor,
actually
added
this
in.
For
us,
it's
something
about
the
its
resources.
The
resources
containers
use
the
way
that
they're
split
up
in
like
the
c
groups,
I,
don't
know
the
details
of
how
it
works.
Okay,
okay,.
D
C
So,
actually
weave
all
this
back.
There
was
a
regression
of
a
that
was
fixed
by
something
else
that
caused
the
regression.
It
was
later
fixed
by
something
else
because
of
regression,
and
we
were
basically
splitting
log
lines
in
journal
D
so
like
when
the
output
of
the
log
to
The
Container
would
come
out.
C
If
it
wasn't
exactly
on
a
4K
boundary,
it
would
split
and
mangle
it
and
they
were
unraitable,
and
so
we
added
this
as
part
of
some
fix
up
for
the
pr
but,
however,
like
so,
if
you
can
see
this
commit
that
I
pasted
into
the
tab.
That
was
the
original
issue
against
conma
that
we
had
vlogged.
C
B
Remember
related
to
this
as
the
3D
mini
box.
Do
you
mean
this
issue
oops
this
one.
D
C
Right,
yeah,
that
was
part
of
the
fakes
because
they
were
changing
the
way
they
were
handling
logging.
B
To
yes,
Meanwhile,
my
search
for
yeah,
my
concern
was
just
basically
to
be
clear
about
what
is
the
effect
of
disabling
this
this
flag.
If
we
give
the
options
to
the
user,
the
containers
not
only
for
one
of
them.
C
B
C
There's
regression
introduced
by
blank,
which
is
a
fix
done
by
myself,
which
was
to
fix
an
regression
done
by
atom
in
a
different
PR,
and
then
they
have
yeah.
So
all
the
way
down,
there's
finally
like
a
workaround
PR
and
so
and
that's
where
the
C
group
split
was
introduced,
I'm
going
to
change
the
way
we
handle
the
logging
through
Journal
B
to
containers.
C
A
A
C
A
Yeah
I
do
remember
this
mangled
Vlogs
thing
though
yeah
it
just
starts
printing
like
new
lines
everywhere
and
stuff.
So.
B
Disabling
this
may
may
break
logging.
C
C
A
Yeah,
maybe
we
can
drop
it
I'm
about
to
like
I,
guess,
look
really
closely
at
what
versions
We
support
I,
don't
know
what
we'll
do
it
for
Quincy
like
a
little
different
backboard
dropping
it,
but
this
is
like
we
used
for
some
slightly
older
os's
there
that
might
have
some
older
podman
versions,
especially
we've
got
just
got
fixed
last
year,
but
maybe
for
like
R.
We
can
go
look
at
this
see
if
this
is
even
necessary.
A
Yeah
that
doesn't
really
change
I.
Think
in
my
eyes
the
the
need
for
this
current
workaround
thing
at
that
pull
request
to
have
to
be
configurable,
because
this
guy
is
trying
to
use
Quincy
and
he
just
can't
deploys
osds
at
all,
so
I
mean
like
you
can't
I,
don't
know,
there's
not
really
much
else
to
do
about.
We
have
to
sort
of
let
them
turn
it
off,
but
I
don't
know
if
we're
necessarily
ready
to
say
I
always
drop
it
now
either
like
it
seems
like
we
have
to
at
least
investigate
more.
A
But
what
I
kind
of
wanted
to
do
was
get
that
patch
in
and
get
it
back
ported
really
quickly.
So
that's
he
could
have
it
fixed
for
16,
2
or
17
to
4
or
whatever.
The
next
thing
is
it's
coming
out
supposed
to
come
out
soon.
B
A
D
C
A
That
comes
back,
I,
guess
the
original
reason
I
sort
of
post
this
year,
because
I
figured
this
for
this
fix
is
coming
I'm
going
in
anyway,
just
because
yeah
we
need
it.
I
want
to
get
it
for
this,
this
guy
for
the
conceive
release
but
I'm
almost
like
of
how
do
we
handle
these
in
general,
like
if
we
need
to
do
this
in
the
future,
like
there's
some
other
reason
like?
A
Oh
but
there's
another
problem,
we
now
need
to
have
this
other
flag,
but
then
this
other
flag
is
now
broken
on
this
distro
and
that
they'd
rather
them
have
this
other
problem
than
have
the
thing
be
totally
broken.
This
one's
need
to
be
configurable
as
well
believe
this
could
happen
again,
and
then
we
just
have
to
just
keep
doing
this,
or
we
just
add
in
these
flags
everywhere
that
we
could
eventually
also
drop
them.
A
We
could
remove
the
clock
at
some
point
and
the
constant
option
and
all
that
probably,
but
it
is
sort
of
a
I,
don't
know
I'm
talking
about
with
a
nice
way
of
handling
this
like
generically,
like
we
like
to
give
you
some
way
to
like
to
say
like
oh
just
don't
put
this
thing
in
so
they
can
customize
it
a
bit
but
I.
Don't
it's
harder,
it's
easier
for
when
they
adding
new
things,
but
it's
hard
to
say
the
generic
way
to
remove
a
flag
or
something
like
that.
That's
that's
tricky.
C
Yeah,
do
we
happen
to
know
what
version
he
was
using
on
Rocky
Linux,
because
this
is
something
we've
that's
why
we
have
that
huge
Matrix
before
I
mean
it
kind
of
got
into
the
same
struggle
of
how
much
of
the
podline
interface
should
we
be
working
around
because
that
becomes
nightmarish
after
a
while,
too
and
I.
Think
that's
why.
A
Yeah
I
remember
the
whole
The
podman
Matrix
thing
yeah,
so
I
guess
podman
is
4-1-1,
I,
don't
know
if
he
says
which
we
probably
ask
him.
He's
responded
to
me
before
when
I
posted
in
there.
C
This
is
at.
A
A
Yeah
so
I
don't
know
it
was
this
whole
period
where
I
feel
like
we
just
kept
discovering
weird
podman
bugs
and
you've
seen
a
couple
of
them
like
you,
you
looked
at
that
tracker.
I
know
before,
like
the
pro
Cooks
thing
yeah.
So
there's
a
bunch
of
things
like
that,
and
we
were
like
all
right.
This
podman
version
actually
works.
I.
Remember
we
had
like
try
to
like
in
pathology,
pulling
certain
repos
to
have
certain
podman
versions
for
our
testing,
so
they
would
all
pass.
A
But
this
this
Matrix
is
very
like
outdated,
I.
Think
at
this
point,
because
it
only
goes
up.
It's
like
1621
is
the
latest
version
we
talked
about
on
here
and
when
I
thought
about
five
minutes
to
3.0
like
this
guy's
using
like
4.1.1,
it's
just
totally
outside
of
The
Matrix
I
guess
we
should
probably
eventually
add
more
roads
to
it.
Saying
like
it's,
4.1.1
works
for
17.2
onwards
or
something
I'm
gonna
have
to
check
what
we're
actually
testing
right
now
in
tautology.
Maybe
we
can
add
some
more
to
this.
A
Yeah,
that's
a
side
topic,
but
I,
don't
know
we
bring
this
up
again.
I
guess
we'll
bring
up
the
the
stuff
again,
but
I
was
I
was
very
happy
to
hear
this
like
this.
Is
there
any
good
way
to
handle
something
like
this,
where
we
have
this
this
generic
sort
of
flag
that
we
need
to
add
in
for
some
reason,
then
it's
broken
for
certain
people
and
we
wanted
to
configurable.
Can
we
make
that
a
generic
way
rather
than
having
to
have
a
specific
conflict
option
for
every
single
time?
This
happens.
A
I'm
not
sure
there
actually
is.
That
might
be
too
hard.
C
A
Yeah,
that's
kind
of
where
I
worried
about
because
I
even
thought
about,
like
maybe
like
they'd,
have
some
sort
of
list
of
things
like.
Oh,
don't
include
this
and,
at
the
very
end,
we'll
like
block
all
those
but
every
time
I
think
about
these
things.
I'm,
like
oh
there's,
so
many
ways
to
them
like
break
this
and
like
all
the
impossible
debug
like
which
flag
did
you
remove
and
which
flag
did
you
add
or
whatever.
B
I
totally
solved,
but
so,
instead
of
having
like
a
config
options
for
separate
Flags,
we
have
a
config
option,
which
is
the
command
line
we
feed
to
podman.
So
this
way
anybody
can
put
whatever
they
need
and
we
don't
need
like
to
start
adding
and
removing
configurations.
B
A
And
just
if
it's
something
like
this,
where
it's
like
it's
broken
for
everyone,
like
you,
kind
of
want
it
to
be
like
by
default
on
so
because
otherwise
we
got
to
like.
Basically,
we
were
doing
this
release.
We
have
to
then
document
like
hey
like
we
want
you
all
to
go
and
add
this
flag
in
yeah
that
doesn't.
D
Scale
for
users,
I,
agree,
I.
Think
the
other
thing
is
it
going
back
to
the
original
issue.
Did
he
have
this
problem
on
all
his
systems
or
it
sounded
to
me,
like
you
were
saying,
one
of
his
systems
was
hitting.
D
A
A
Yeah
I
also
thought
about
the
same
thing,
because
I
have
no
idea
he's
the
first
one.
I've
ever
heard
of
who's
actually
ran
into
this,
but
it's
just
like
it's
so
hard
to
debug.
This
sort
of
thing
like
I,
don't
know,
I
thought
if
we
just
had
this
conflict
option
for
him
I'm
like
all
right.
It's
the
thoughts
on.
If
no
one
touches
it
everything's
the
same,
but
for
this
one
guy
he
can,
he
can
turn
it
off
or
I'm
sure
there
won't
be.
There
should
be
at
least
one
other
guy.
C
A
D
Other
thing
we
could
think
about
in
the
future
is:
if
you
need
to
do
this
command
line
thing,
but
it
could
be
on
a
per
host
basis.
You
could
tweak
the
system
D
unit
file
to
inject
args
from
like
a
file
like
an
environment
file
that
would
be
like
a
lower
level,
debugging
thing
and
that
wouldn't
clutter
all
like
the
cephadium
and
orchestrator
module
code.
It
would
just
be
like
hey
take
this
environment
variable
and
just
slap
it
onto
the
end
of
the
podman
command
line,
kind
of
thing.
A
Yeah,
that's
why
I
do
a
manually,
debugging
things
I'll
go
and
like
add
things
like
that
right,
we
could
have
something
that
did
four
people
like
it'll
go.
Add
this
to
the
unit.run
file
and
I'll
restart
it.
The
only
issues
we've
run
in
with
those
is
that,
if,
for
whatever
reason,
the
manager
module
ever
wants
to
restart
that
demon
or
redeploy
it
whatever
it'll
just
override,
all
of
that
it
goes.
D
Away
yeah:
that's
why
I
said
like
a
file.
If
you
had
a
like
Etsy,
you
know
system,
you
know
so,
if
adm.overrides
EnV
or
something
and
put
crap
in
there,
I
don't
know
I'm,
definitely.
B
Yeah,
something
which
is
residents
in
the
host
not
related
to
the
files
inside
the
container.
A
Yeah
definitely
an
idea,
it's
just
a
matter
of
I
guess
how
do
we
format?
That
way,
because,
like
say
for
this
exact
case,
we
have.
We
are
passing
this
one
flag
and
we
want
to
remove
that
flag
like
how
do
I
specify
that
in
my
file
like
in.
D
This
case
I
think
we
should
double
check
it,
but
in
this
case
I,
since
you
can
do
like
Dash
Dash
I
had
the
man
Page
open,
you
can
do
dash
dash
whatever
it
is.
D
D
groups
equals
I
think
if
you
specify
it
a
second
time
like
c
groups
equals
enabled,
which
is
the
default
Behavior
it'll
just
like
override
the
first
invocation.
That's
no
good
for
dash
of
e
options,
which
are
like
additive
but
I
think
for
an
option
like
this.
It
might
be
fine,
but
we
could
fiddle
around
with
it
and
test.
If
that's
the
case,.
A
We
could
try
it
guys
if
there's
a
way
to
do
that,
even
if
that
works
before
you
talk
about,
where
you
just
add
this
new
c
groups
thing
in
that,
maybe
he
could
actually
get
around
this
with
the
extra
container
arms
already
yeah.
D
A
Yeah
I
think
it
does
them
last
sorry
eight
hours
ago,
but
I'd
have
to
verify
that,
but
I
can
at
least
check
it
and
I
can
tell
the
guy
about
it
and
see
if
it
if
it
works
right.
If
it
fixes
that-
and
we
can
scrap
this
old
PR
I-
don't
need
this
anymore.
Now
I
was
working
work
around,
but
I'll
give
that
a
try,
I
haven't
even
thought
of
that
I
didn't
know.
You
could
do
that.
You
could
like
re-specify
the
flag,
because
it
would.
D
A
D
A
A
That's
right
all
right,
yeah
we'll
do
that!
Then
I'll
I'll
check.
If
the
c
groups
enabled
the
works.
B
The
command
line
is
not
long
if
you're,
considering
having
this
command
line
as
option,
would
save
a
lot
of
this
kind
of
situations
without
any
any
new
code
from
our
site.
B
The
the
department
yeah
command
line
right
command
line
so
having
all
the
command
line
as
as
a
config
option.
So
by
default
we
put
whatever
we
need
and
then,
if
some
user
needs
to
add
it
or
to
remove
or
to
do
whatever
with
the
command
line,
then
he
can
do
on
his
own
risk.
A
That's
kind
of
what
Mike
was
saying
before
if
we
could
have
some
option,
that
that
does
the
whole
thing
up
for
you
like
that,
and
then
they
could
change
it.
We
could
pass
that
down
and
another
thing
that
would
that
would
maybe
work,
but
it
seems
like
very
dangerous
and
very
ugly
and
I.
Don't
know.
B
And
you
know
nobody
will
or
nobody
should
play
here's.
D
B
Right,
like
the
Linux
kernel,
if
you
want
to
boot
with
some
specific
option,
you
just
modify
the
the
command
line
and
the
input
with
this
option.
D
A
Well,
if
there's
a
good
way
to
know
how
to
check,
maybe
I
need
to
find
an
old
version
and
try
to
do
that
and
see
if
it
destroys
my
logs,
so
I
can
find
out.
D
B
A
D
It
upstream
and
he's
using
an
unsupported
distro.
He
should
be
at
least
willing
to
try
things
I'll.
A
A
And
then
we
had
a
lot
of
strange
ideas
about
how
we
could
handle
those
in
the
future,
but
I
think
we'll
we'll
leave
that
for
another
time,
if
it's
necessary
yeah.
A
A
We're
good
that
case
we'll
end
here
and
I.
Guess:
I'll,
probably
mostly
people
in
here
on
the
stand
up
on
Thursday,
yes,
yeah.