►
From YouTube: 2020-01-13 :: 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
Yeah
yes
and
start
welcome
to
today's
orchestra
meeting.
Let's
start
with
safe
IDM.
A
A
But
it
was
basically
out
of
the
question
that
the
self,
a
DML
orchestra
that
needs
to
properly
validate
things
before
he
is
going
to
or
before
safe
alien
monster,
is
going
to
remove
any
other
STIs,
that's
obvious.
But
who
is
going
to
be
responsible
for
draining
our
T's?
And
there
was
no
real
conclusion,
because
that
turns
out
that
the
dashboard
recall
is
kinda
the
the
wrong
core
to
discuss
this
and
yeah
that
that's
here
again.
B
B
So
I
mean
if
you,
if
you
have
a
look
at
the
full
picture
you
basically
so
there
is,
there
is
ZERO.
The
steps
are
in
the
design
document.
But
to
reiterate,
you
basically
set
the
way
to
zero.
If,
if
you
want
to
be
really
really
paranoid
right
and
then
you
wait
for
the
OST
to
be
empty
and
that
needs
to
be
happen
in
some
kinda
in
the
kind
of
loop
fashion
and
then
you
need
to
stop
it
and
do
other
things,
that's
probably
outside
of
the
scope
of
the
module,
but
you
so
in
the
past.
B
We
we
figured
out
that
having
lots
of
different
options
for
draining
an
OSD
and
for
accept
or
for
the
acceptance
criteria,
is
it
well?
Let
me
rephrase
it
so
when
you,
when
you
want
to
remove
more
or
as
these
than
just
one,
you
can
utilize
this
okay
to
destroy
or
okay,
to
stop
feature
from
self
right.
And
if
you
pass
it,
if
you
pass,
it
was
the
one
two
three
four:
it
will
maybe
tell
you.
B
Yes,
it's
it's
fine
to
destroy,
and
but
if
it's
not
fine
to
destroy,
you
basically
have
to
remove
only
one
OSD
before
or
well
drain
it
and
for
the
data
to
become
available
or
redundant
on
on
other
oddities.
First
before
you
remove
the
other
ones,
and
that
logic
tends
to
get
a
bit
more
complex
and
if
we
implement
that
in
safe
ATM
also
Brooke
needs
to
implement
that.
C
B
D
A
D
B
I
mean
since
it's
potentially
a
long-running
process,
at
least
having
some
kind
of
progress.
Reference
is
certainly
helpful,
but
the
rest
is
totally
up
for
discussion.
I
mean
it
kind
of
depends
on
how
complex
the
operations
and
the
modes
will
be,
that
this
thing
can
operate
in
I
mean,
if
you
just
say,
drain
and
then
a
set
of
oils
DIDS.
There
is
probably
only
be
one
query,
and
that
is:
is
it
done
or
not
right
or
what?
Which
percentage
is
this
done?.
D
B
D
B
So
that
depends
on
your
cluster
setup.
I
think
we
utilized
that
okay
to
destroy
command
and
that
will
basically
tell
you
how
many
power
or
mites
it's
more
or
less
I,
think
the
okay
to
stop
command
and
because
it
will
essentially
is
stopping,
will
essentially
start
draining
and
OSD,
and
so,
if
you
pass
or
if
the
okay
to
stop
command
yields,
okay
for
the
OSD
IVs
that
you
passed
in,
you
can
just
drain
them
altogether.
B
However,
if
the
set
of
OSD
is
that
you
want
to
drain,
are
not
okay
to
stop
or
to
destroy.
You
have
to
take
a
bit
more
conservative
approach.
Just
you
know
they
drain
half
of
them
or
something
that
would
say
the
impact
on
your
cluster
depends
on
on
multiple
different
things.
First
of
all,
of
course,
in
your
configuration,
how
much
you
even
allow
to
allow
the
recovery
to
take
up
your
your
bandwidth
and
also
on
your
cluster?
E
Question
one
question
is
about
I:
think
that
you,
yes,
what
I
think
that
you
have
talked
about
to
have
a
new
stated
model
in
order
to
implement
a
this
kind
of
operations
and
to
use
this
Orchestrator
from
this,
this
architecture
model
from
others
for
the
camera,
for
example,
from
the
book
or
case
later,
model
of
death,
SSH
or
AVM
model.
This
is
that
true,
is.
Is
that
the
the
manager
main
idea
I.
B
Haven't
really
thought
about
where
exactly
this
thing
would
live?
Okay,
if
it
turns
out
to
be
an
Orchestrator
focus,
trader
module
or
a
manager
module
that
put
the
ideas
so
the
main
purpose,
wherever
this
thing
lives,
is
to
reuse
code
for
all
the
orchestrators
for
rope,
stuff,
idiom
and
and
anthem,
okay,.
E
Okay
is
because,
when
I
think
that,
maybe
in
this
case,
what
we
are
going,
but
what
we
are
trying
we
are
trying
to
do
is
just
to
to
add
a
new
operation
to
the
FFA
estate.
I
am
ok,
so
probably
to
implement
the
new
safe
OST
drain
operation
in
the
FF
questi
diamond
could
be
useful
because
what
I
think
that
he
he
went
to
to
put
the
right
flat
the
red
flags
in
order
to
do
not
reuse
this
device
and
to
provide
information
about
the
progress.
B
Not
sure
because
if
you
implemented
in
the
stuff
was
the
demon
I
mean
it
would
be
obviously
the
the
lowest
point.
But
if
you
want
to
pass
multiple
LSD
IDs
in
it,
you
somehow
need
to
aggregate
it
aggregate
the
data
that
you
get
from
each
of
the
virtual
Damon.
So
something
needs
to
be
on
top
of
the
daemon.
Yes,.
E
Yes,
but
to
execute
operations,
but
several
quest
is
probably
should
be
implemented
in
each
of
their
status,
but
coding
directly
to
the
day
operation
that
each
OS
is
the
elastic
should
be
able
to
go
to
execute.
That
is
to
to
drain
the
touristy
and
and
to
provide
information
that
is
safe
to
remove
physically
removed
in
the
diversity
is
what
I
mean
the
the
operations
with
a
lot
of
horses
or
that
involve
even
different
different
nodes.
Different
host
I
think
that
should
be
implemented
and
I.
E
Think
that
is
logical
is
sensible
to
to
think
that
this
kind
of
things
should
be
a
sort
in
Turkish
state
orderly,
but
each
okay
stator
can
solve
this
in
different
way.
Okay,
because
in
this
case,
what
we
are
going
to
have
his
only
containers
but
at
containers
that
are
managed
managed
in
different
different
phone.
So
I
think
that
promise
is
what
idea
to
implement
a
single
OST
during
operation
for
each
in
his
a
FF
was,
did
I
ignore.
B
I
think
so,
I
think
a
drain
operation
in
itself
doesn't
really
like
some,
because
you
could
either
set
it
out
or
set
their
way
to
zero,
which
is
essentially
draining
it,
and
there
is
also
this
I
mean
they'd.
The
okay
to
stop
and
okay
to
destroy
comments
are
already
part
of
the
policy
Damon,
so
the
manager
module.
Would
you
really
just
be
there
to
aggregate
data
and
wait
for
results
and
give
you
better
better
data
back
that
you
can
work
with
in
different
orchestrators.
A
A
B
B
A
A
Okay,
then,
let's
postpone
the
pies
and
package
the
picture
when
stages
here,
where
the
cure
failure
I
had
a
cure
for
that
today,
no
inventory
found
on
host
smithy
one
three
two
at
was.
A
A
A
A
A
Provide
good
errors
regarding
SSH
connections
like
if
you
have
wrong
as
the
edge
configuration
like
like
problem
with
the
keys
or
stuff
like
that
Union,
you
will
always
get
a
host
not
found
errors
by
excitement
which
is
not
really
helpful,
and
but
that's
about
it
right
now.
The
other
potential
problem
is
that
you
are
going
to
have
a
scaling
problem
with
exit
exact
neck,
because
everything
is
synchronous
and
the
only
way
to
paralyze
executions
is
to
have
a
thread
pool
a
kernel
thread
pool
so
that
that's
not
immediate
an
immediate
problem.
A
But
what
what
do
you
think.
A
A
B
It
doesn't
really
create,
creates
a
problem,
but
the
usability
really
suffers,
I
think
yep.
Definitely
if
you
I
mean
currently,
we
only
have
like
or
whatever
three
to
four
mold
clusters,
but
and
reality
people
have
more
than
hundred
and
here
in
there
a
note
goes
down
and
if
that
happens,
and
she
you
happen
to
the
query
device
LSO
service
allows
for
something
it
will
just
bail
out
and
tell
you
that
your
ssh
configuration
is
somehow
wrong.
Instead
of
telling
you
that
that
the
oh
sit
down
that
kind
of
yep.
A
A
A
E
E
A
Let
let
me
that's
it
installation,
and
then
here
we
have
this
set
this
documentation
part
which
curls
the
safe
ATM
binary
from
the
gate,
source
repository,
which
has
the
unfortunate
implication
that
the
let
the
executable
needs
to
be
exactly
the
same
as
the
file
we
are
well
be.
The
executable
needs
to
be
exactly
the
same
as
the
source
code,
which
means
we
cannot
aggregate
multiple
plasma
modules
into
one
executable
and
I
have
a
pull
request
out
there
to
improve
that
situation.
A
E
I
don't
understand
what
the
reason
for
me
is
not
I
think
that
we,
who
have
the
same
name
in
this
equitable
and
one
file,
I,
think
that
in
in
Python
there
are
several
solutions
in
order
to
establish
a
Couture's,
so
I
think
that
not
not
a
good
reason,
I
think
the
opinion.
A
part
of
that
well,
I,
think
that
in
this
moment
probably
it
can
be
something
that
will
be
easy
to
manage.
E
A
Painful,
it's
already
painful.
If
you
look
into
these
notes
code
of
the
ferry
and
then
you
will
see
lots
of
steps
of
stuff,
that
would
be
already
part
of
multiple
modules.
Like
we
have
an
implementation
for
locking
and
implemented.
We
have
a
command
that
interface.
We
have
auxiliary
laboratories
and
everything
is
crammed
together
in
one
big
source
file.
That's
not
the
best
way
to
structure
a
project,
that's
going
to
be
maintained
for
a
long
time,
so
we
have
to.
We
really
have
to
improve
the
situation.
There
review
my
pull
request.
A
B
E
E
E
But
even
this
anyway,
that
what
probably
there
is
a
request
about
one
day
requirement
in
order
to
start
in
15
OneNote
is
to
have
access
this
node
throat
SSH
and
you
need
to
copy,
probably
appear
in
the
in
the
remote
node.
You
should
what
Africa
at
requirement
to
have
staff
and
to
have
people
start,
and
if
you
have
these
requirements
is
basically
the
same
and
when
we
are
not
in
telling
with
something
that
we
know
that
is
going
to
if
more
put
lot
of
problems
in
now
in
hand
in
the
future.
E
A
A
A
E
B
Frankly,
they
only
needed
once
right;
they
only
needed
once
for
bootstrapping
purposes
and
after
that
the
manager
takes
care
of
copying
it
over
to
to
the
target
host.
Yes,
oh,
if
it's,
if
it's
a
one-time
thing,
then
that
can
even
be
I,
don't
know,
distribute
it
as
whatever,
as
whatever
is
what
we
can
stall
the
easiest.
You
can
even
like
distribute
it
as
a
snap
or
whatever
I.
Don't
know,
know
that.
A
C
G
Yeah
so
as
part
of
a
big
change
to
introduce
the
role
mode
on
for
overseas
on
PVCs,
I
I've
refactored
the
way
we
were
discovering
devices
as
well
as
the
way
we
were
validating.
Ost
over
discs
are
good
candidates
to
become
OS
DS
by
relying
on
the
simple
image
which
recommend
which
sent
me
to
append
a
bunch
of
issues
against
cell
volume
so
that
we
can
improve
the
command
in
the
process.
I
also
realized-
and
we
have
been
discussing
this
for
weeks
now-
that
we
were
not
supporting
partitions
and
I.
G
Don't
think
we
had
a
good
reason
why
we
just
had
that
really
stupid
check
where
we
just
keep
the
device
if
we
detect
a
partition-
and
we
don't
have
any
reasons
to
do
so
so
now.
What
we
do
is
that,
if,
instead,
if,
if
we
detect
that
they
are
like
filesystem
signatures,
we
don't
do
anything,
but
we
shouldn't
be
doing
it
sit
for
a
minute
or
should
be
doing
it.
But
I
have
an
issue
open
for
that.