►
From YouTube: Ceph Orchestrator Meeting 2020-08-31
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
First
topic:
I
have
one
again
now
that
we
last
week
we
had
a
meeting
about
nfs
integration
in
the
dashboard
and
in
the
volumes
module
to
to
give
you
a
quick
summary
about
what
what's
the
problem
there.
We,
unfortunately,
we
now
have
two
incompatible
nfs
exports
management
implementations,
so
the
the
first
nfs
management
was
developed
for
nautilus
in
the
dashboard.
A
A
Both
are
kind
of
based
on
the
same
so
the
volumes
nfs
code
base
is
based
on
the
dashboard
nfs
code
base,
but
now
I
they
they
have
diverged
to
a
point
where
we
no
longer
can
use
both
nfs.
A
A
So
my
friend,
just
engaged
in
that
versa
is
engaged
in
that
and
beefer
is
also
engaged
in
the
discussion.
A
B
A
A
A
Not
really
incompatible
but
yeah,
they
they
can't
work
both
on
the
same
nfs
export.
A
They
have
a
different
feature
set
and,
as
far
as
I
know,
they
they
corrupt
export.
The
other
module
created.
A
A
A
Yeah
that
affects
rook
as
well,
because
the
dashboard
supports
creating
nfs
exports
with
rook
and
the
volumes
nfs
module
also
supports
writing
nfs
exports
with
work,
so
it's
brook
rook
and
civilian
are
both
affected
by
this
problem
that
we
have
right
now.
Okay,
next
topic,
extraceph.conf
entries
there
there
seems
to
be
a
problem
that
the
theft.conf
generated
by
config
general
generate
minimal.
A
Conf
is
not
complete
to
an
extent
that
is
needed
by
every
different
edge
case
like,
for
example,
things
that
need
to
be
set
before
the
initial
monitor,
connect,
set
up
and
then
dennis.
I
think
also
mentioned
to
me
that
some
blue
store
settings
need
to
be
loaded
before
the
monitor
connection
is
I
I
don't
know,
I'm
that's
a
bit
above
my
area
of
expertise,
so
I
came
up
with
basically
two
solutions.
A
One
is
to
provide
a
way
to
extend
the
self.conf,
which
is
deployed
to
all
daemons
in
the
cluster.
A
A
Yeah,
I
hope
we
can
get
both
things
merged.
One
temporary
solution
for
having
a
flexible
way
to
do
everything
right
now
and
the
other
one
for
properly
generating
a
minimal
self.conf.
B
B
Okay,
I
think
some
of
the
options
might
only
take
effect
on
my
boot.
A
A
In
general,
everything
should
work
using
the
monster
and
if
you
really
want
to
debug
a
specific
setting
or
if
you,
if
it
turns
out
that,
then
that
one
specific
setting
really
must
be
set,
then
you
can
use
this
kind
of
extra
sev.conf
setting,
but
but
then
you're,
basically
on
your
own.
A
A
Hopefully,
at
least
at
least
that's
a
plan
or
the
idea
or
in
in
general,
it
should
be
automated
and
the
user
shouldn't
need
to
mess
with
a
theft..
A
A
A
B
B
A
A
Next
topic
pit,
one
in
container
so
right
now
we
don't
have
a
proper
build
one
in
containers,
which
means
that
the
safety
month
itself
are
pit
one
which
is
not
the
way
rook
is
doing
things.
Rook
uses
tiny
right
now.
Safe
adm
does
nothing
which
leads
to
basically
two
problems.
A
Treats
processes
with
pit
1
as
special
and
doesn't
create
cordon,
so
it
works
as
long
as
the
car
dump
was
not
created
by
by
pit
one
but
another,
but
but
a
different
pit.
Then
you
automatically
are
getting
columns
so
thankfully,
docker
and
seth
idm
putman
supports
a
flag,
called
minus
minus
init,
and
that
should
work.
A
A
C
Thoughts
just
one
thing:
I
think
that
maybe
we
should
implement
this
as
the
default
option.
Okay,
I
think
that
it
works
in
suse.
It
works
in
rail,
eight,
okay
and
well.
Probably
there
is
a
problem
with
centos
and
fedora,
but
it's
something
that
could
be
solved:
okay,
because
it's
just
the
star
dependencies.
So
what
I
propose
is
to
include
a
node
in
the
installation.
C
A
It
seems
to
be
a
general
of
a
packaging
problem,
because
I
mean
the
functionality
just
doesn't
work,
so
it
seems
to
be
that
the
the
inner
process
is
just
missing
from
the
requirements
of
the
ipm.
A
Another
idea
to
automate
this
would
be
to
to
first
spawn
the
first
monitor
with
minus
minus
init
and
see
if
it
fails
and
if
not,
then
silently.
A
C
C
So
this
is
something
that,
in
this
moment,
is
fixed
in
some
operating
system
and
in
others,
is
not
the
solvent
okay.
So
I
think
that
probably
well,
the
detection
in
the
in
the
operating
system,
where
it
is
not
fixed,
is
to
include
this
dependency
in
order
to
make
bot
man
work
with
all
the
parameters.
So.
C
That
is
is
not
in
our
field.
Okay,
what
we
can
do
is
just
to
to
warn
the
user
that,
if
you
are
going
to
use
this
operating
system
with
this
person,
you
know
that
portman.
He
has
this
requirement
and
you
need
to
install
it
in
order
to
have
properly
configured
postman
and
the
possibility
to
have
core
files
and
the
problem
of
the
of
the
clean
of
the
zombie
containers.
A
I
I
really
would
like
to
get
this
pull
request
in
and
getting
merged
getting
it
merged,
even
disabling
it
by
default
and
then,
following
up
with
a
new
pull
request
and
that
prints
out
a
warning
or
prints
out.
B
C
B
Yeah
numbers
like
I
guess
that
the
point
is,
I
think
we
do
yeah.
You
want
to
be
able
to
have
core
jumps
in
by
default
regardless
of
the
mechanism,
and
where
is
the
distribution.
C
C
About
this,
this
issue,
okay-
and
this
is
because
in
in
both
places,
the
catatonic
package
is
not
distilled
okay,
so
it's
a
requirement
that
probably
poor
man
needs
the.
I
think
that
is
something
that
could
be
solved
by
podman
by
in
in
this
case,
what
is
happening
is
that
in
trail,
eight,
for
example,
we
have
in
red
hat
back
that
is
close
in
this
moment
that
includes
this,
this
package
and
for
success
in
the
case
of
centos,
ubuntu
and
fedora.
The
thing
that
is
not
solved
yet.
A
We
have
a
different
option.
We
have
a
different
option.
Instead
of
making
it
the
default,
we
could
make
it
a
hard
requirement
that
this
is
working
and
then.
C
Met
okay,
so
make
the
make
a
check
when
we
are
creating
the
the
buda
strap
cluster
in
order
to
see
if
it
is
started
or
not,
or
it's
time
that
we
are
going
to
deploy
new
diamonds
in
in
in
new
horse.
A
C
A
A
I
mean
for
downstream:
that's
not
a
big
issue
because
for
both
downstream
solutions,
that's
that's
going
to
be
made
anyway,
because
we
really
need
a
column,
that's
mainly
about
accessing
upstream
clusters,.
C
A
A
B
A
Okay,
then,
let's
go
with
making
it
a
hundred
coin,
not
requirement
next
topic:
config
templates,
so
we
in
in
surf
idiom.
We
have
a
few
config
templates
for
for
for
generating
rafana
configs
for
generating
alert
manager,
configs
stuff,
like
that
and.
A
They
are
hard
coded
and
there
is
basically
no
way
to
properly
modify
that
in.
In
any
case,
we
could,
in
theory,
make
those
ginger
templates
over
writable,
using
the
the
config
key
store.
A
Like
you're,
pushing
a
new
ginger
template
as
a
config,
key
blob,
intercept
and
then
the
next
time
the
3dm
deploys
the
daemon
it's
going
to
use
a
new
ginger
template.
The
the
overwritten
ginger
template
benefit
is
obviously
that
it's
going
to
be
more
flexible,
but
there
are
a
few
downsides
with
it.
C
It
is
going
to
be
needed
in
the
moment
that
you
try
to
install,
for
example,
another
type
of
time.
There
is
a
full
request
about
installing
general
containers
or
something
like
that:
okay
yeah,
it
could
be
used,
for
example,
to
install
it's
a
proxy
okay
and
keep
alive
in
order
to
to
make
a
rcw
endpoints.
A
A
So
a
use
case
would
be
if
you
really
have
the
need
to
modify
the
alert
manager
configuration
to
to
match
what
you,
what
you
really
need,
which
which
contains
stuff
that
is
not
supported
by
the
federal
or
you
need
to
to
modify
the
grafana
configuration
template
to.
I
don't
know,
do
do
what
you
really
need
to
do,
which
is
impossible
right
now.
C
C
Yes,
yes,
it
has
the
the
high
degree
of
flexibility
and
also
the
the
high
risk
of
doing
the
things
in
the
in
their
own
in
their
own
way.
So
I
think
that
it's
a
good
idea
to
to
include
the
cincia
templates.
Okay,
that
is
something
that
will
help
us
in
the
future
to
to
make
things
more
easy
to
incorporate
new
diamonds
and
new
well
and
to
to
make
easy
to
customize
configuration
files.
C
Okay,
because,
probably
you,
you
are
not
going
to
need
to
know
all
the
internal
variables
that
we
have
in
favor,
but
it
it
provides
to
the
user.
I
think
that
flexibility
in
order
to
configure
the
diamonds
in
the
way
that
he
wants.
C
A
So
you,
you
would
need
to
derive
a
new
container
image,
replace.
A
Which
is
interesting,
I
I
thought
that
the
the
sense
would
be
to
just
don't
do
that
and
please
know,
and
but
it's
it's
interesting
to
to
see
actually
some
interest
in
this
one
and
having
something
like
this:
okay,
yeah,
that,
let
me
think
about
it,
and
maybe
we
can
come
up
with
a
very
simple
solution.
A
A
A
A
Next
topic,
it
turns
out
that
you
can
now
actually
call
talks
without
any
parameters
by
default
in
the
pi
bind
in
the
source
pipeline
manager,
folder
and
execute
all
the
self-adm
and
orchestrator
and
step
schedule.
A
And
the
last
topic
I
I
had
in
mind-
was
refractoring
refractorization
of
the
self-ident
binary
how
we
want
to
go
forward
with
this?
How
do
we
want
to
start.
C
Yeah,
well,
I
think
that
we
probably
the
the
point
where
the
start
is
trying
to
to
make
simple
in
the
bootstrap
command.
Okay,
if
I
think
that
is
the
key
point:
okay,
because
basically
is
the
main
functionality
and
what
we
have
is
a
big
sequence
of
different
steps
involving
different
dimensions,
different
different
kind
of
things:
okay,
so
probably
if
we
can
refactor
the
the
bootstrap
command
using
classes.
Okay
for
the
different
elements
that
we
are
managing
in
the
bootstrap
command
is
something
that
is
going
to
to
make
more
readable.
The
the
fidm
code.
A
A
That's
even
worse,
it's
even
worse,
worse
deploying
a
new
sav
cluster.
If
you,
if
you
follow
the
option,
documentation.
A
C
Yeah,
so
it
seems
that
we
need
what
the
the
fifth
step
in
order
to
do.
Good
refactorization
using
modules,
should
be
to
enable
the
packets
more.
C
C
C
Prior
to
start,
when
I'm
going
to
prepare
a
house
or
to
check
if
a
host
is
ready
to
start
the
3dm,
the
first
thing,
one
of
the
things
that
we
can
do
in
the
same
way
you
have
proposed,
for
example,
to
check
if
the
catatonic
package
is
installed
or
not
in
order
to
use
the
minus
minus
in
the
parameter.
Okay,
why?
We
can't
check
if
a
fadm
package
is
standard
in
order
to
see
if
the
host
is
is,
is
is
possible
to
include
this
host
in
the
in
the
class,
then.
C
Okay,
but
we
are
not
talking
about
the
bargain-
we
are
talking
about
production
systems
and
and
to
to
have
possibility
to
do
things
in.
I
think
in
the
in
the
more
easy
way,
and
one
of
the
problems
that
we
have
is
that
we
are
using
something
that
is
a
file
that
you
can
get
from
anywhere
okay,
so
this
is
a
problem.
It
is
it
to
to
to
be
sure
that
when
we
start
fidm,
we
are
installing
this
version
from
this
place.
Okay,
this
is
the
the
fifth
problem
to
solve.
C
A
No,
no,
no!
No!
No!
That's
a
problem
having
self-ed
m
installed
on
the
whole
host
only
solves
one
problem
and
that's
the
initial
deployment.
The
bigger
problem
is
that
we
can't
use
a
package
mode
when
doing
upgrades
right
now.
A
A
If
we
really
want
to
want
to
want
to
go
forward
with
the
packet
mode,
then
we
have
to
make
sure
that
the
cli
between
the
self
area
manager,
module
and
the
self
adm
package
is
compatible,
plus
minus
two
major
versions,
and
then
the
the
the
api
between
the
safety
and
binary
and
ceph
itself
is
also
compatible
plus
minus
two
meter,
major
versions,
which
is
a
tremendous
requirement
that
we
don't
fulfill
right
now.
A
It's
not
yet
there
and
that's
the
problem
that
you
can't
use
the
package
upgrades
right
now.
C
So
the
problem
starts
if
we
start
to
define
a
basic
set
of
of
commands,
that
we
are
going
to
need
in
order
to
make
up
upgrades
inside
the
fidm
and
make
this
set
of
commands
it.
C
Okay,
not
all
the
the
ap
of
the
vdm,
only
part
of
the
of
the
ap
and
with
this
part
fix
it
in
order
to
to
use
this
part
to
make
the
upgrades
we
can
do
that,
but
the
important
thing
is:
that
is
something
that
in
this
moment,
we
are
implementing
all
the
functionality
with
all
the
ap
okay
and
what
we
have.
We
are
worried
because
in
the
future,
when
we
are
we,
we
will
need
to
to
change
the
version
to
update
to
a
newer
version.
C
The
ap
is
not
going
to
be
compatible.
We
are
worried
about
that.
Okay,
yes,
but
but
we
are
worried
that
the
new
things
and
the
new
implementation
that
we
are
going
to
do
is
not
going
to
to
follow
the
design
principles
and
the
requirements
that
we
have
marked
in
this
moment.
C
C
Okay,
so
it
is
difficult
because
we
need
to
to
think
very
well
what
what
is
going
to
be
this
minimum
set
and
what
is
the
commands
and
the
parameters
that
we
are
going
to
need
it's
difficult
to
do
that,
but
it
is
something
that
we
should
do
and
after
that,
what
the
only
thing
that
we
need
to
do
is
not
to
break
not
to
break
this
in
the
future
and
for
not
to
break
this
in
the
future
is
is
to
have
a
version
in
the
ap
and
and
to
follow
the
rules.
I
think.
A
And
in
order
to
have
a
safe,
adm
minus
minus
version,
we
need
to
get
rid
of
of
users
curling
directly
from
the
github
repository,
because
that
prevents
anything
that
that
really
breaks.
A
If
it
really
prevents
every
change
to
the
source
code
between
developers
and
users
like
adding
a
proper
minus
minus
version.
So.
A
If
we
really
want
to
do
the
package
mode,
then
we
we
need
to
have
a
defined
stable
api
between
the
fadm
and
ceph
and
the
manager,
module
and
self-adm,
and
for
that
we
really
depend
on
proper
totality
tests.
B
A
B
B
A
B
A
A
It
I
mean
the
the
version
typically
also
contains
the
github,
the
git
hash,
which
would
be
impossible
to
add.
B
Instead
of
that
being
an
unstable
location,
making
that
stereo
location
doesn't
change.
B
A
A
And
in
general
I
really
want
to
get
rid
of
downloading
users
directly
from
github.
B
B
I
do
think
it
is
important
to
preserve
that
kind
of
that's,
not
packaged
mode
of
installation.
That's
the
simplest
way
to
get
running
with
a
single
command
yep.
A
A
Problem
is
there,
there
are
distributions
where
we
don't
have
proper
packages
like,
for
example,
ubuntu.
There
is
no
f
adm
package
because
we
don't
release
it.
B
Yeah
and
it's
just
an
extra
step
for
users
to
do
which,
whether
you
try
focusing
on
removing
steps
rather
than
adding
them.
B
The
initial
this
is
like
the
first
thing
is
someone
totally
new
to
stuff
might
see
and
and
want
to
make
that
as
frictionless
as
possible,.
A
C
B
C
Why
could
we
cannot
publish
public
package
for
theftm.
C
A
A
Yeah
here
and
we've
talked
about
that
in
january
january,
already
chemical
yeah.
C
A
Here,
that's
that's
a
list
of
things
that
problematic
when
just
making
a
package.
A
Oh,
we
are
beyond
that
already
we're
close
to
6
000
lines
of
code,
python,
blob,
okay
and
lots
of
classes
within
this.
So
I
I
really
think
that
we
really
should
also
focus
on
making
it
a
proper
python
package
before
even
considering
making
it
making
the
mode
default
to
the
package
mode.
C
A
C
Okay,
the
the
third
step
is
to
create
a
well
to
to
to
investigate
to
to
to
collect
what
is
the
ap
that
we
are
using
in
order
to
make
updates
and
put
a
totality
test
in
order
to
verify
that
this
is
not
changed.
Okay,
probably
we
will
need
to
to
to
reach
an
agreement
about
what
is
the
minimum
set
of
commands
that
we
need
to
to
update
and
what
parameters
we
are
going
to
use
and
with
that
we
will
have
a
free
dm
in
in
package
mode
okay.
C
So
that
is
another
thing
that
we
need
to
do.
That
is
in
the
in
the
moment
that
you
are
using
theft,
adm
and
well
fdm
binary
when
you
are
using
from
the
state
or
the
fedm
command,
we
are
using
some
things
in
root
mode
that
are
not
used
in
packets
mode
okay.
So
we
need
to
check
if
this
code,
how
to
how
to
change
this.
This
program?
Okay,
so
basically
these
four
steps-
or
maybe
I
can
send
you-
welcome
right.
A
A
That
I'm
just
proposing
that
we,
let
me
split
it
up
in
in
different
steps
and
instead
of
having
one
big
refactorization,
that
does
everything.
Instead,
we
go
step
by
step
fixing
one
problem
at
a
time.