►
From YouTube: 2019-09-30 :: Ceph Orchestration Meeting
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
So
we
come
to
do
this:
okay,
with
a
call
yeah
the
again
other
day.
A
A
3600
3x3
yeah
just
have
a
look.
It's
about
the
part
that's
going
to
be
executed
on
localhost
if
we
split
out
def,
safe,
deploy
into
the
low
clouds
part
and
the
the
remote
execution
button
that
progress
is
about
any
local
hotspot,
yeah
for
look,
I've,
reworked
my
access,
client,
Python,
pull
request
and
yeah.
Let's
see
how
it
goes.
A
So
the
thing
is:
I
really
need
some
kind
of
static
type
checking
in
manager
in
the
rook
module,
because
otherwise
we
simply
have
no
way
to
make
sure
that
the
API
that
the
manager,
rook
module,
expects
matches
the
one
exposed,
Barack
and
just
doing
sevens
and
J's
and
stuff
really
does
not
work.
If
we
want
to
have
it
working
properly
and
I
love
was,
you
would
need
to
have
enormous
test
coverage,
two
hundred
percent
minimum
so
having
every
edge
case
being
being
executed.
A
That's
right
now
kind
of
unrealistic,
so
we
really
have
to
resort
to
some
static
type,
checking
and
yeah
that
pull
request
a
temporary
test
and
does
this
but
makes
use
of
Brookes
F
giant
Python
library
which
I
did
but
has
no
home.
Yet
that's
rook,
the
rook
SEO
3
5,
4,
3
and
yeah.
What
would
be
nice
to
get
this
into
the
rook
organization
or
into
the
self-organization
or
be
part
of
the
project
or
be
part
of
the
Ceph
project
yeah?
By
the
way?
B
B
B
B
A
A
A
B
A
C
A
B
A
A
A
A
D
B
B
B
D
D
B
D
Because
the
last
last
issues
they
they
had
apparently,
is
that
the
pod
restarts,
though
the
container
restarts
and
so
yeah.
Somehow
the
IP
we're
distributing
in
in
the
config
file
for
the
USD
was
different
than
the
IP
of
the
container,
so
that
the
OSD
couldn't
actually
listen
on
that
IP
because
it
wasn't
I
mean
because
it
was
in
existence,
was
an
existing,
so
I
guess
if
they
use
once
they
will
use
one
one
to
this.
This
should
probably
go
away.
D
B
Because
he
he
said
that
originally
he
had
this
F
config
overrides.
While
he
was
running
one
dot
Oh,
but
then,
as
part
of
the
upgrade
he
removed
those
settings.
The
upgrade
mentions
that
it's
recommended
to
remove
those
custom
settings
and
use
the
new
Mon
config
settings.
It's
think
he
did
that
and
then,
but
since
he
didn't
expect
the
overrides
to
be
needed
or
even
be
picked
up,
move
them
are
saying
that
your
fix
yeah,
because
I
think.
D
Even
though,
even
though
he
was
using
the
config
from
the
months
then
who
at
this
point,
the
OSD
was
still
using
an
old
self
config
file,
which
was
wrongly
generating
IP
addresses
so
as
part
of
my
change
I'm
not
using
that
config
file
anymore,
so
that
the
OSD
would
directly
fetch
whatever
is
in
the
Mon
can
be
stored
because
that
because
the
set
off
takes
precedence
over
the
the
config
store,
that's
why
they
were
having
issues
too
so
yeah.
As
you
said,
it's
a
bunch
of
unrelated
problems,
cascading
effects
and
so
forth,
but
yeah.
B
D
D
B
D
B
D
D
Well,
actually,
we
we
do
have
recommendations
once
you
set
limit,
and
the
problem
is
that
if
you
set
the
limit,
then
the
code
will
reject
some
of
the
limits
you
might
be
putting
in
place
because
they're
too
low.
So
that's
really
really
recommended
for
like
production
environments,
if
you're
running
productions,
then
they
have
to
run
limits.
It's
really
I
mean
it's
highly
encouraged.
Obviously,.
B
D
B
D
Yeah
I
have
a
batch
for
us
or
for
the
external
Chester
again
bit
more
around
the
lines
of
how
six
hath
improvements,
though
relaxing
the
restrictions-
yes,
and
actually
removing
some
of
these,
like
a
necessary
code
to
coming
from
the
old
lot
so
old,
that
fig
override
type
that
we
introduced.
And
we
moved
right
away.
D
A
B
C
Well,
we
have
two
possibilities
of
implementation
here:
okay,
one
is
able
to
do
the
big
job
I'm,
trying
to
move
the
discovery
that
I
am
or
to
what
the
discovery
that
I
am
a
prophet
is
that
a
task
board
needs
or
that
a
Turkish
trade
or
disrespecting
okay
and
well,
there
is
another
that
is
more
more
easy.
Okay,
I
have
seen
that.
C
Basically,
the
difference
is
that
we
have
between
the
output
that
we
have
in
the
discovery
diamond
and
what
doctor
stated
aspects
is
basically
the
available
okay,
and
this,
this
field
is
done,
is
composed
of
four
or
items.
Okay,
different
items.
What
I
have
seen
is
that
when
maskull
e
is
for
items
are,
one
of
them
is
if
this
is
thermally,
removable
or
not.
Okay,
again
that
we
can
get
this
information
from
the
hardware.
The
third
time
in
the
screen
I'm
on
the
other
one
is
the
autonomy
is
already
present
in
this
moment
in
the
in.
D
C
Limit
or
e
the
other
one
is
locket.
That
is
what
is
a
operation
that
is
doing
that
volume
and
basically,
what
is
to
win
is
to
try
to
walk
in
to
open
in
in
screw
Sigma
the
device
in
order
to
see
is
it
if
it
is
rocket
or
not
my
PI
other
process?
Okay,
so
probably
I,
don't
know
this
is
well.
There
is
another
one
that
is
a
cassette
by
theft,
disk
and
I.
C
C
Okay,
if
we
implement
the
local
check
in
the
the
discovery
diamo,
we
can
provide
exactly
the
same
information
that
the
state
or
needs,
and
there
is
no
need
to
do
any
kind
of
new
new
element
or
new
it
item
in
the
third
part
of
the
orchestrator.
In
order
to
fulfill
these
requirements.
I
don't
know
if
it
is
ok
or
not.
I
think
that
is
the
more
easy
solution,
not
obviously,
if
something,
if
something
changed
in
the
podium
I,
probably
we
will
need
to
check
again
this
madness.
A
So
my
idea
is
I
have
two
ideas.
One
is
that
we
could
modify
the
orchestrator
data
structure
to
be
to
match
what
safe
volume
inventory
prints
out
that
we
are
not
inventing
a
new
data
structure
within
the
orchestrator
but
reuse.
The
existing
data
structure
is
exposed
by
cell
for
you
and
the
other
idea
would
be
to
just
remove
rook
out
of
the
loop
and
just
pass
through
the
output
of
safe
volume
to
the
orchestrator.
A
So
that's
the
one
one
direct
path
from
chef
volume,
on
on
the
logos
on
the
system,
a
path
through
path
through
the
orchestrator
and
make
it
available
for
the
despot
to
properly
generate
drive
groups,
because
the
drive
group
generation
generation
is
basically
if
you
are
in
order
to
display
what
the
user
wants
to
do.
You
have
to
do
the
exact
same
algorithm
as
the
volume
when
applying
drive
groups.