►
From YouTube: 2020-04-13 :: 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
A
B
C
D
A
C
With
so,
we
said
to
people
okay,
well,
it's
just
to
to
comment
about
what
is
the
current
state
of
hydrograph
resilience
the
health
of
the
of
the
system
when
something
happens
in
the
in
the
manager
and
the
possibilities
to
recover
in
the
manager
with
def
ADM,
okay,
I'm
a
little
bit
worried
about
that?
Possibly
is
not
it's
not
heavily
tested.
Okay,
I
think
that
what
this
this
is
not
going
to
happen
with
with
proof,
because,
basically,
what
we
have
a
Hinda
basis
square
methods?
C
Okay,
so
this
kind
of
things
are
really
solve
it
in
most
of
the
cases,
but
with
Def
ADM
I
think
that
it
could
be
interesting
to
to
have
a
view
of
what
is
the
current
state?
Okay,
Oh,
let's,
let's,
let's
store
they
date
them
and
I.
Think
that
we
can
discuss
in
the
in
the
next
meeting
with
a
with
a
upstream
people.
B
C
Well,
the
the
only
thing
that
we
have
in
Fairfield
Elam
is
that
you
can
deploy
to
manages
okay,
but
it's
not
clear
that
this
is
going
to
work
or
I
think
that
we
should
test
more
heavily
this.
This
kind
of
cost
analysis
in
in
general
in
general,
I
think
that
what
this
is
I
think
that
is
something
that
is
logical.
Okay,
if
FA
diem
is,
is
ready.
New,
okay
and
probably
when
this
kind
of
things
about
the
health
of
the
system
and
and
about
to
have
safe
a
DM
working
in
disaster
recovery
situations.
C
D
It
is
a
Stratos
I
mean
what
happens
today.
You
know
I
mean
so
in
each
of
the
managers
we
have,
you
know
said
you
know.
If
ADM
there
are
four
components
that
they're
you
know
dashboard
requires.
So
we've
got,
you
know
stuff
ADM
we've
got
Prometheus,
we've
got
the
alert,
client
or
whatever.
We
call
it
and
then
grow
fauna
I
mean
so.
D
Do
we
replicate
each
of
those
and
every
single
manager
node
in
the
end,
do
we
use
H
a
proxy
to
fill
those
over
UPS
I
mean
that's
the
ideal
solution
right
I
mean
you
know
from
the
point
of
view
of
what
we
would
like
to
see.
I
would
think
at
least
from
speaking
it
from
a
downstream
perspective.
What
actually
happens
they
upstream
I
mean
I
brought
this
up
last
week,
I,
don't
think
you're
at
the
meeting,
Travis
and
I
just
said.
A
Before
for
a
rook,
specifically,
we
only
have
a
single
manager
and
we
don't
even
allows
start
any
more
than
one
manager
that
the
you
know.
We
rely
on
kubernetes
to
restart
the
manager
or
move
it
to
another
node.
If
it
fails
so
yeah,
there's
only
one
at
a
time
and
when
it
starts
up-
and
we
just
expect
the
modules
to
come
up
wherever
it
wherever
it
comes
up
and
I.
A
D
You
have
the
rook,
you
have
the
rook
implementation,
which
could
potentially
be
different
than
as
you
were
talking.
Then
the
you
know
the
non
rook.
You
know
def
ADM
derivative
right
inside,
but
we
think
that
would
require
those
multiple
stacks
with
an
H,
a
foxy
implementation,
I
mean
I,
I
would
think,
but
I
mean
you
know,
that's
really
would
be,
would
be
nice
to.
You
know,
get
our
hands
around
and
you
know
try
to
figure
out
how
we
could
model
the
orchestration
layer
they
handle
both
and
be
effective.
D
You
know
who
are
less
the
same
at
the
higher
levels,
work
the
same
and
then
at
the
underlying
you
know
rock
or
Steph
ATM
implementation
do
the
appropriate
things.
You
know
there.
So
I
don't
know.
That's
my
you
know
my
thought
on
it,
but
I
have
no
idea
what
we're
planning
on
where
we're
going,
and
you
know
typically,
do
we
typically
just
increment
in
a
way
there
or
do
we
talk
about
it?
Architectural
II
figure
it
out.
I
mean
you
know.
How
do
we
typically
sit
because
I
haven't
paid
a
lot
of
attention
upstream?
D
D
A
B
C
Yes,
I
feel,
then
that
we
have
two
different
things:
one
is
to
have
several
managers
working
together:
okay,
and
that
is
possible.
Okay
and
another
different
thing
is
the
failure
of
it
and
the
disaster
recovery
situations.
Okay,
because
in
this
moment
what
we
have
interphase
am
basically
is
something
that
is
only
the
project,
one
time
and
using
system.
C
Okay,
this
is
one
possibility.
Another
possibility
is
just
to
restart
the
service
in
in
another
in
another
node
and
doing
this
darkus
later
I.
Think
that
what
my
opinion
is
that,
probably
we
are
putting
too
much
things
in
the
Indian
state
or
larger
in
faith.
I
am
with
this
I
prefer,
for
example,
the
distribution
that
Jeff
has
planted,
reducing
it's
a
it's
a
proxy
and
to
have
the
possibility
to
have
several
managers
working
together.
C
B
I
mean
if
you
were
just
fighting
failure
and
moving
in
the
active
one
to
a
different
vendor.
That's
what
the
manager
already
does
today
by
itself,
if
you're
running
multiple
of
them
I'm
tact.
If
the
active
one
goes
down,
another
one
become
the
active
one
and
quite
a
standby
mode,
essentially,
and
it's
kind
of
coordinated
by
the
monitor,
having
an
active,
enter
controlled
by
them.
Out
of
the
wrapper
I'm
telling
the
managers
which
which
manager
is
supposed
to
be
active.
So
it
seems
like
in.
B
C
Yeah,
but
what
I
don't
understand
with
that
is
why,
for
example,
intro
is
not
possible
to
do
that.
Ok,
what
I
think
that
sage
taught
us
that
it
was
not
going
to
work,
because
we
are
because
the
two
managers
are
going
to
try
to
expose
that
the
same
service
with
the
same
port.
I.
Don't
remember!
Why
wait
what
what
was
the
vision
of
that
yeah.
A
Two
managers,
one
of
them,
is
active
and
one
of
the
passive
if
you
put
a
service
endpoint
in
front
of
it
like
a
cheap
proxy,
and
it
just
does
round-robin
routing
great
you're
not
going
to
get
the
same
manager
every
time.
But
you
want
to
point
to
the
active
one
right.
So
how
do
we
point
to
the
active
one
from
that
service
endpoint,
but
a
normal
kubernetes
service?
Well,
it
sort
of
acts
like
a
proxy
right
just
round
robin.
C
C
C
A
C
B
Yeah
I
can
make
a
lot
of
sense
to
have
that
to
like
the
manager
and
geology.
Suite
is
already
like.
We
call
it
like
writers
that
kill
randomly
killed
like
it
was
DS
monitors
could
add
a
similar
one
for
the
measure.
Yeah.
D
But,
but
actually
when
you
failover,
though
you
know
trying
to
maintain
state
I
mean
it
might
pass
what
lives
I
mean.
We
never
really
tried
to
maintain,
say
it's
a
very
difficult
problem
right,
so
you
would
just
at
least
have
a
another
place.
Another
point
to
go,
get
into
a
dashboard
and
we
manage
again
and
go
over.
You
know,
go
back
to
where
you're
at
you
know
versus
I
mean
I,
don't
think
I
mean
you
guys
are
thinking
that
we
could
actually
maintain
state
and
pick
up
the
command
where
we
left
off.
D
D
A
Well,
I,
guess
and
I
regard
I
mean
rook,
doesn't
work
itself,
isn't
really
stateful,
but
where
we
rely
on
because
we're
not
to
use
a
net
CD
underneath
that,
basically
to
remember
what
our
desired
state
is
and
then
every
time
the
every
time
real,
quick,
sup
or
the
operator
runs
or
whatever,
then
we
go
and
make
that
desired.
State
happens,
then,
what's
a
feed,
yeah
I'm
following
a
similar
model
right
where
there's
some
desired
state
and
then
try
to
apply
it.
A
D
C
A
C
C
A
B
D
But
what's
interesting
in
that
that
model
as
well
but
I,
you
know
again,
I,
don't
know
what
we
do
today
upstream,
but
do
we
do
we
put
all
the
components
we
require?
You
know
within
each
you
know
within
each
manager.
You
know,
that's
the
that's
the
question.
I,
don't
know
if
we
do
like
refine
I,
don't
think
we
do,
because
if
you,
if
we
do
fail,
overs,
we
don't
get
grifone
or
right.
So
that
indicates
to
me
that
we
didn't
have
multiple
instances
Gravano
downloaded
and
again.
D
Is
that
something
we
want
to
do
upstream
and
just
bring
it
downstream
with
us
and
have
a
work
or
that's
something
we
just
want
to
do
downstream.
I,
don't
really
care,
but
it
would
be
a
good
good
discussion
to
have
the
more
we
do
upstream
and
we'll
consistent
and
the
more
it
gets
tested.
The
more
you
know
the
better
off
we'll
be
for
the
long
run
so
but
it'd
be
nice
to
just
understand
what
those
are
the
four
components
down.
You
know
loaded
on
each
one
and
I.
Guess
you
won
me
and
Danny.
C
C
C
Well,
not
too
much
things
from
my
back
I
think,
but
that
is
a
requisition
truck
out.
The
integration
test
with
theft
money
is
ongoing.
Okay,
you
can
take
a
look
okay,
I
will
I
would
like
to
to
have
it
integrated,
okay
or,
for
example,
to
create
a
new
suite
in
order
to
run
this
street
every
night
or
even
a
separate
street
okay,
because
what
I
want
to
have?
Is
this
intelligent
test
running
all
the
nights
against
the
remastered
version
of
F?
C
Okay,
in
order
to
detect
as
soon
as
possible
any
problem
with
a
with
a
a
PE
or
with
there
is
little
interface?
This
moment
is
only
covered
in
debates.
You
said
use
cases
or
is
this,
and
basically
all
the
commands
to
check
a
status
device
status
at
this
status?
This
kind
of
things,
okay,
but
I,
think
that
it
could
be
interesting
to
have
his
running
as
soon
as
possible
in
order
to
to
1s.