►
From YouTube: 2020-05-18 :: 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
That's
not.
We
have
a
few
topics
already
on
the
list.
Oh
the
yurts
are
talking
that
15
are
two.
The
two
is
going
to
be
released
soon
very
soon,
hopefully
today
other
than
that.
Maybe
in
this
week
not
fully,
but
it's
John
I
know
that's
the
scuffs
of
topic,
streamlined
CLI,
oh.
C
A
C
Porticus
that
use
subversion
at
three
five
or
six
fork.
We
are
actually
using
the
same
foot
path
now
for
a
demon
ad
and
for
effect,
apply
with
the
only
difference
that
we
do
not
store
the
spec
right.
So
it's
some
sort
of
a
one-shot.
But
besides
that
is
I
would
say.
99%
the
same,
so
we
could
drop
it
and
add
a
parameter
to
apply
that
it's
called
at
no
save
non-persistent,
one
shard
or
whatever.
That
signifies
that
the
user,
that
this
is
this
is
a
one-time
operation
and
that
you
know.
A
Actually
I'm
doing
it,
the
other
I'm
doing
it
the
other
way
around
with
three
five
or
six
four
I'm
changing
it
so
that
demon
ad
we
use,
apply
you
swim,
but
it's
nothing
to
do
one
shot
or
not.
Nothing.
A
non,
persistent
or
Nothing
relates
to
nothing
really
to
no
save
it's.
It's
just
SAS
them.
Esper
assistance.
A
C
C
C
Basically,
try
to
come
up
with
a
concept
that
is
more
centered
around
the
back
pain,
because
that's
a
central
building
block
of
this
of
this
organ
of
this
whole
walkers
everything,
because
you
basically
control
your
entire
cluster
with
Beck's,
right
and
so
respect
as
the
object,
and
then
you
have
a
verb
which
is
most
most
likely,
the
operation
that
you're
running
it
on.
So
we
have
like
a
less
which
is
an
alias
to
self
focus
to
LS.
This
could
be.
C
Apparently,
when
we
want
to
modify
something
our
workflow
is
that
we
export
it,
then
change
it
and
then
reapply
it
and
it's
kind
of
tedious
right.
So
if
you
have
something
that
shows
you
your
specification,
then
you
should
also
have
a
way
to
modify
your
visit
oftentimes.
You
only
want
to
change
one
parameter
and
it's
less
scary.
If
you
only
touch
one
thing
and
yeah,
oh
and
the
next
one
is
just
a
modified
example
for
account.
I
want
to
cement
modify
example
for
unmanaged.
C
Then
there
is
Specter
plier,
which
is
the
same
thing
as
our
currently
existing
apply.
It's
revamped
a
bit,
and
then
we
have
to
preview
the
in
which
maybe
deserves
its
own
sub-command,
maybe
not
I'm,
not
sure,
but
since
every
spec
can
have
a
preview,
it
is
non
non
exclusive
to
oppose
these.
This,
then
this
is
pretty
interesting
information
which
maybe
should
deserve
its
own
yeah
and
then
pros
and
cons.
The
only
cons
that
I
could
really
find
this
D.
C
So
the
most
annoying
thing
is
that
the
transition
period
that
you
have
to
take
the
old
commands
around
for
these
four
articles
and
then
it's
a
bit
more
verbose,
which
is
you
know,
a
con
because
it
type
more.
But
it's
also
pro
because
it
actually
reads
like
an
English
sentence,
which
I
always
found
a
bit
find
a
bit
more
convenient.
Always
oh.
C
C
The
only
one
one
feature
request
is
that
it
will
apply
and
also
that
it
applies
to
the
current
apply
and
by
function
we
should
detect
if
there
is
a
as
a
spec
with
with
an
existing
service
name,
and
we
should
not
just
overwrite
because
that
I'm
a
scary
to
detect
this
primitive
and
then
you
know,
reject
it
or
whatever
anyways.
That's
it:
okay,
okay,.
D
I
think
that
is
a
very
good
idea
and
maybe
probably
we
should
I
will
try
to
see
what
magnetics
is
to
in
in
this
moment.
Okay,
with
the
apply
common
in
order
to
manage
theories,
and
you
see
the
same-
the
same
approach
because,
for
example,
I
think
that
well
shown
or
later
we
are
going
to
need
things
like
affinity
or
anti
affinity
and
more
things.
So
the
spec
file
is
going
to
become
more
complex.
C
A
E
D
Know
even
who
is
doing
that?
Okay,
something
now
with
cpad
okay
in
order
to
provide
develop
air
and
clean
the
environment.
Okay,
but
this
is
another
possibility,
is
just
to
comment.
If
this
could
be
better
I,
don't
know
it
is
just
to
try
to
use
the
local
folder
there
for
the
sources
for
the
source
of
forecast.
A
toes
in
in
the
containers
running
the
mana
here,
I,
don't
know
if
a
is
could
be
a
good
idea,
or
we
start.
This
is
better
than
this.
I
don't
know
that.
A
Let
me
go
yes,
sir.
Yes,.
D
D
It's
about
it's
a
proxy
implementation.
I
have
been
what
they
can
I
look
to
the
current
code
that
we
have
in
Devastator,
and
what
I
have
seen
is
that,
basically,
we
are
repeating
exactly
the
same
approach
or
it's
the
diamond,
okay,
and
in
a
way
that
I
think
that
they
probably
to
do
to
pass
to
a
more
object-oriented
approach,
could
be
well
to
simplify
code
and
make
things
more
more
easy
to
understand
and
and
to
execute
and
to
and
to
maintain.
D
Okay,
basically,
is
what
I
think
that
we
can
see
that
very,
very,
very
easy
in
lot
of
the
function
that
we
have
in
FF,
Orchestrator,
okay,
one
of
them
is
the
the
ugly
ugly
side
with
function,
but
we
have
this
kind
of
approach
when
we
are
a
numeric
enumerated
things.
Basically,
okay
and
I
think
that
it
could
be
interesting
to
evolve
in
this
direction.
I
see
that
is
okay,
fantastic,
okay
is
everything
down.
D
D
A
To
have
with
specific
implementations
in
the
modulite
PI,
basically
keeping
the
module
dot,
pi
independent
of
the
existing
of
the
actual
service
and
move
all
these
other
specific
parts
in
into
the
services
yeah.
That's
that's
ago,
okay,
at
least
that's
my
goal,
but
we
are
not
there
yet
if
you
have
any
real
factorizations
and
please
do.
D
Yes,
what
I,
what
I,
have
how
this
is
just
to
try
to
implement
it?
And
you
know
it's
a
proxy
service
with
a
nice
new
approach.
Okay,
obviously
I
want.
It
could
be
I
think
that
it
is
going
going
not
to
to
to
be
perfectly
fit
perfectly
with
the
current
code,
but
I
think
that
it
could
orient
us
in
the
in
the
residence.
D
D
Next
step
is
about
is
easiest
that
I
have
seen
that
we
do
not
have
documentation
about
theft
of
a
device
like
on
okay,
so
it's
something
that
we
are
doing
in
the
stated,
but
probably
that
the
commentation
the
right
place
in
order
to
put
this
information
is
in
in
devices
devices
finance
me
it's
just
to
want
to
have
some
advice.
If
this
is
the
the
right
location
and
I
will
I
will
do
the
modification?
No
problem
with
that.
F
Yeah,
hey
guys
yeah,
so
I
started
work
on
drive
group
parsing
in
cell
phone
to
unblock
Blaine
and
yeah.
There's
a
well
first
shot
at
it
in
a
branch
we
might
posted
the
link
and
yeah
since
there's
some
some
open
topics
here,
I
wanted
to
bring
those
up
and
I
mean
name
is
the
name
is
currently
awful,
but
I'm
sure
we'll
find
something
better.
Very
soon
question
is
mostly
or
I.
Think
I
discuss
topic
is
mostly
post
matching,
which
lane
would
like
to
push
down
to
cell
folium,
but
I.
F
C
F
C
E
So
my
I
think
our
our
thought
was
that
if
we
use
that
same
logic,
there
can't
be
any
mismatch
in
like
what
hosts
get
information
applied
to
them
from
I,
like
user
interface,
do
actual
implementation
perspective.
Yes,
like
it's.
A
volume
uses
the
same
like
Python
code
that
dashboard
manager
uses
for
that
kind
of
user
feedback.
So
that
was
that
was
our
reasoning
for
wanting
to
have
the
host
name
matching
done
at
the
set
volume
level.
E
Otherwise,
we
will
probably
grab
a
like
standard
glob
library
from
like,
go
and
use
that
in
work
and
so
I
I
think
glob
machinist
fairly,
well-defined,
but
there
there
is
some
risk
of
like
mismatch,
and
if
that
happens,
then
it
becomes
about
it
was
kind
of
hard
to
track
down
like
where,
where
is
the
you
know,
is
the
bug
in
shaft?
Is
it
then
that
that
implementation
of
Gotham
is
an
incorrect?
Is
it
something
else?
E
So
that's
my
reasoning
for
kind
of
wanting
to
push
it
in
to
set
volume.
It
isn't
actually
necessary,
although
it
does
also
mean
that
rook
doesn't
have
to
like
inspect
the
right
groups
that
it
gets.
It
just
has
to
pass
them
down,
otherwise,
definitely
inspect
them
and
like
see
if
they
actually
have
a
hostname
key
and
like
if
they
don't
that
a
lot.
F
F
E
C
I
would
remember
to
do
this
matching
in
the
orchestrator
modules,
from
your
master
machine,
because
then
you've,
one
single
source
of
truth,
and
it
knows
the
host
names
and
you
can
just
resolve
the
hosts.
I'm
sent
the
drive
trips
to
the
matching
hosts
and
then
just
you
know,
go,
and
if
you
do
the
evaluation
on
the
host,
you
lose.
You
lose
a
lot
of
functionality,
for
example,
reporting
because
you
don't
know
what
will
happen
if
you're
interested
in
that
kind
of
information
and
stuff.
Oh
I,.
E
I
mean
I
I,
think
I,
think
that
that
logic
works
for
the
stadiums,
but
that
logic
doesn't
I,
think
we're
quite
as
easily
for
rustics.
Orchestrator
module
doesn't
really
do
a
whole
lot.
It
just
configures
Rock
to
go
and
do
things
so
it's
some
Python
code,
that's
changing
configuration
that
will
go
and
it
will
run
they
like
rooks,
a
like
system,
that's
written
in,
go
and
so.
A
E
A
E
F
E
F
E
E
F
E
F
I,
don't
know
I
just
didn't
want
to
get.
You
know,
Bach
down
and
bike
shedding
when
I
started
this
okay,
cool
I
think
that's
the
most
of
what
I
wanted
to
know.
I
will
push
this
yeah.
What.
F
Care
I'm
currently
writing
tests
for
it
sorry
I'm
billing
right
now
here
we
are
yeah,
please
if
you,
if
anybody
would
be
able
to
test
that,
that
would
be
very
helpful.
There's
the
link
to
the
rewrite
yeah.
So
this
basically
enables
you
to
like
all
the
same
batch
command
on
on
a
note
over
and
over
again,
and
it
will
only
redeploy
the
or
it
will
only
deploy
that
the
parts
that
are
not
already
deployed.
Of
course,
yeah.
F
F
Otherwise,
if
there's
anything
up
with
that,
let
me
know
if
it
works
for
you.
Let
me
know
I'm
hoping
to
add
like
the
most
basic
testing
this
week,
and
we
have
a
customer
downstream
to
who
would
like
to
try
this
out
so
hopefully
get
some
some
exposure
for
testing
here
and
then
I'm
happy
to
merge
this
into
master,
as
well
as
into
the
several
branches.
F
Right,
that's
why
I
kind
of
listened
up
earlier
so
for
I
have
another
PR
that
I
want
to
test
with
SEF,
ADM
and
well.
I
ended
up
pushing
a
shaman
built
for
that
which
you
know
is
workable,
but
not
ideal
from
just
from
the
delay
perspective
and
turnaround
times
and
whatnot
I
think
it
would
be
really
handy
if
we
had
a
way
of
just
injecting
some
def
code
into
a
running
container.
Ideally,
no
I,
don't
think
anybody
has
been
much
thought
on
that
yet,
but
would
be
nice
if
we
could
figure
out
something
like
that.
C
F
A
F
F
We
have
downstream
needs
in
the
future.
When
you
know
everybody
has
customers
on
this
and
we're
trying
to
debug
some
some
issue
that
only
one
customers
east
and
would
certainly
be
helpful
if
we
could
kind
of
inject
modify
it
boat
into
containers.
Without
you
know,
tearing
down
a
container
and
bringing
it
up
again,
I.
A
F
And
that
to
Christina
you
know-
and
that's
probably
an
issue
at
least
to
how
things
sometimes
get
it,
but
right
now,
but
even
in
even
without
restarting
the
demon,
there's
I
see
a
use
case
for
this,
where
you
know
are
now
in
with
problematic
OS
DS.
We've
definitely
had
situations
where
we
just
patched
like
the
e
blue
store
table
to
well
I,
think
it's
called
or
something
like
that,
would
like
the
tool
with
which
you
can
inspect
a
local
OSD
store
and
that's
independent
of
the
actual
OSD.
F
Yeah
and
that's
I
mean
that's
kind
of
what
I'm
suspecting
will
happen
anyway
for
for
self
all
you
know,
this
is
as
easy
as
just
copying
some
Python
code
with
compiled
code.
That
might
be
a
bit
more
tricky
anyway.
I,
mostly
are
glad
that
this
is
on
the
agenda
that
people
talk
about
this,
because
I
think
this
hasn't
gotten
that
much
attention
yet
I
know
that
amiss
at
least
looking
at
I
know
debugging
tools
for
for
the
container
world.
E
E
Updated
files
into
a
running
container
and
that
if
it
is
a
binary
and
just
restart
that
binary,
although
at
that
point
you
may
as
well
restart
the
container
as
well
I,
would
think
or
yeah
I
mean
it's
different
for
things
that
are
running
we
restarted
versus.
You
know,
like
even
I,
think
a
Python
process
like
replacing
Python
files
you'd
still
want
to
restart
whatever
container
you
there's
multiple
files
into
just
to
make
sure
that
they're,
the
Python
interpreter
wasn't
still
running
with
the
most
files.
E
But
if
it's
like
configuration
files
or
something
like
that
or
like
for
web
servers,
it's
very
easy
just
to
like
push
new
HTML
information
and
then
like
really
like
a
web
server
reload
that
information
so
I
in,
in
my
opinion,
it's
it
makes
most
to
update
a
built
container
and
then
just
restart
whatever
single
container.
You
want
updated
if
you're,
trying
to
like
really
optimize
for
the
least
amount
of
like
iron
adjourn.
A
Mainly
only
works
for
or
changing
a
safe
ATM
itself
and
in
developing
the
safe
IDM
another
module,
because
the
manager
has
the
nuts
outside
of
the
container
outside
of
the
containers
and
then
it's
just
restarting
the
manager
and
keeping
all
existing
demons
running.
You
know.
So
that's
a
workflow
for
safe
edium
itself
and
unsaved
like.
A
Safe
EDM
knows
how
to
deploy
a
specific
or
to
upgrade
or
to
modify
or
to
redeploy
a
container
with
a
specific
image.
So
you
use
some
def
config
and
then
using
the
k
container,
underscore
image
with
the
correct
demon,
ID
and
then
you're
calling
a
safe,
ODS
redeploy
daemon,
and
then
it's
going
to
get
redeployed
with
the
correct
container
image,
but
that
still
coming
from
the
registry
or
that's
still
coming
from
any
registry,
either
a
local
one
or
yeah.
F
F
B
C
What
is
the
build
time
of
a
container
and
uploading
it
to
your
local
registry?
If
that's
pretty
fast,
though,
wouldn't
at
least
be
terrible,
because
there
is
a
there's.
A
third
scenario,
for
example,
I
encountered
this
one
trying
to
debug
OSD
deployment.
Do
you
can't
modify
the
OSD
code
because
there
would
be
a
new
container,
the
new
container
is
being
spawned,
and
so
the
the
code
that
you
wanted
to
change
must
already
exist
in
the
container
image
that
you
responded.
Oh
yeah.
C
B
B
B
A
B
C
Mean
the
police
versus
a
DM
development,
it
wouldn't
be
as
bad
right.
So
if
we
can
somehow
reduce
the
the
time
that
it
takes
to
rebuild
the
image
container,
air
image
and
yeah
yeah,
you
know
picking
up
the
new
container
image
is
a
matter
of
seconds
and
you
know
turnaround
time
of
minute,
max
I
would
say
and
I
find
and.
A
That's
exactly
what
Josh
is
just
a
Josh
is
going
to
do
he's
going
to
an
instead
of
rebuilding
the
image,
we're
just
linking
or
mounting
the
Pisan
source
code
into
the
running
container,
which
means
that
you're
going
to
have
instant
updates
of
your
codebase
when
we
start
in
containers,
nice
yeah,
that's
kind
of
what
I
thought
would
be
nice
to
have,
and
just
poke
tough
I
think
he's.
He
hadn't
really
time
to
actively
working
it.
But
at
least
at
this
plan.
C
B
F
Well,
thanks
a
lot
all
right:
yeah
I
was
just
going
to
say,
I'm
out
of
things.
I
wanted
to
bring
up.