►
From YouTube: Ceph Orchestrator Meeting 2021-06-22
Description
B
There
is
a
request
ongoing
and
what
what
I
have
seen
is
that
is
possible
to
that.
In
my
my
laptop,
you
will
need
a
what
a
powerful
laptop
okay,
because
in
order
to
set
up
one
master
and
a
couple
of
workers,
you
will
need
enough
drum
memory
and
probably
hard
disk
if
you
want
to
do
a
test,
some
different
tests,
but
what
I
think
that
that
is-
and
it
is
going
to
to
make
it
easy-
also
to
start
to
to
develop
integration
test
also
with
the
coordinates
cluster
and
rook.
A
A
B
Well,
I
I
prefer
kcli
in
that
parliament
is
more
complicated.
Okay,
but
obviously
is
more
mature
at
kcla.
So
every
each
of
the
tools
has
advantage
and
disadvantage.
A
Yeah,
the
interestingly,
if
you
are
having
a
super
complicated
vagrant
file,
then
you
are
coding
ruby
and
if
you
have
a
complicated
kcli
plan,
then
you're
doing
things
with
ginger
2
templates
I
mean:
does
it
really
matter
right,
either
you're
programming
in
ginger,
2
or
you're
programming
in
vagrant
in
in
ruby?
That's
that's
how
it
is,
but
I'm
not
really
opposed
to
it.
B
C
A
A
Yeah
a
few
a
few
seconds
ago,
I've
talked
with
melissa
about
perimeko
and
I
would
like
to
bring
it
up
to
the
to
the
group.
Here.
A
A
They're
doing
some
they're
doing
very
in-depth
ssh
option
parsing
with
some
reduced
functionality
on
top
of
what
ssh
can
do
right
now
and
we
are
already
hitting
those
limitations.
A
So
is
paramecus
really
the
way
to
go
for
us.
We
are
we're
already
seeing
a
few
issues
where
the
where
we
have
to
pass.
A
The
logging
output
we
have
to
capture
the
logging
output
of
parameter
because
parameco
also
is.
E
I've
heard
about
it,
but
I
haven't
looked
into
it
because
I
thought
we
needed
a
synchronous
library,
but
if
async
works
as
well,
then
I
can
check
it
out.
I.
E
On
the
link
that,
on
the
link,
it
says
that
partial
support
for
open,
ssh
style
configuration
files.
A
D
A
I
mean
yesterday
we
we
came
up
with
a
an
ssh
library
that
as
a
prototype,
that
is
about
100
lines
of
code,
that
does
everything
we
need
more
or
less.
A
It
just
wraps
the
call
to
ssh
right
yeah
and
I'm
wondering
if
we
need
a
library
at
all
if
we
can
just
wrap
it
in
100
lines
of
code.
A
An
input
and
an
input,
string
and
executes
a
command
on
the
remote
host,
because
you
can
only
specify
a
single
command
when.
A
A
And
that
easily
starting
to
get
super
cumbersome,
because
you
would
have
to
wrap
every
single
command
that
you
execute
on
the
remote
host
in
a
tiny
python
script
function
in
a
best
best
script
function,
an
alternative
would
be
to
execute
a
python
interpreter
on
the
remote
host
and
then
execute
a
tiny
python
script.
A
To
me,
I'm
not
I'm
not
super
sure
what
we
want
to
go
forward
with,
but
no
that's
why
I'm
bringing
this
up
here
in
this
group.
Here
I
mean
for
me
the
cleanest
way
of
doing
things
is
to
bootstrap
the
ssh
connection,
using
a
10
line,
or
maybe
20
line
piece
of
python
and
then
create
a
or
execute
a
remote
python
function
for
every
command
that
we
do
in
order
to
get
the
result
back
to
the
to
the
original
host.
A
Yeah,
I
guess
it
can
do
that
the
remote
ssh
might
be
an
option
for
us.
Indeed,.
A
Yeah,
I
think
so
too
I
mean
we
are
doing
a
lot
of
calls
that
are
individual
chords,
like
the
vdmls
f80m.
A
Networks,
cfdm
check,
host
fidm,
get
effects
and
having
to
create
a
new
ssh
connection
for
every
command
that
we
execute
is
super
slow
on.
On
the
other
hand,
if
we
are
going
full
into
the
agent,
then
we
would
probably
be
okay
with
having
it
non-persistent.
A
A
It
was
back,
then,
when
we
decided
to
use
remote
versus
parameco
when.
A
Noah
yeah,
when,
when,
when
noah
created
the
ssh
orchestrator,
the
question
was
remoto
versus
perrymaker
and
alfredo
was
really
against
parameco,
and
I
think
he
was.
B
What
about
to
use
ssh,
or
example,
nico
just
to
start
to
make
the
initial
operations
with
a
remote
host?
And
we
are
talking
about
to
have
a
fifth
agent
in
each
of
the
nodes?
What
about
to
implement
a
tfpc
protocol
between
the
different
nodes
in
the
cluster.
A
A
Morning
that
fpc
protocol
can
even
be
an
http
get
request.
So
it's
it's.
We
don't
need
to
implement
a
new
protocol.
We
can
just
extend
the
agent
the
the
future
agent
functionality
to
also
deploy,
demand
and
stuff
like
that
means.
A
Yeah
we
just
have
to
trigger
it
right,
and
everything
else
goes
via
the
typical
rest
api
that
we
are
having
a
pull
request
already.
Thankfully,
so
we
don't
need
to
implement
some
kind
of
crazy
way
of
doing
things
we
just
need
to
trigger
it.
A
Yeah,
but
but
extending
it
to
also
do
deploying
things
should
be
not
super
complicated.
I
guess
if
we
can
get
rid
of
ssh
at
that
point,.
A
But
we
still
have
to
deploy
the
the
agent.
A
But
that
also
doesn't
feel
that
great
to
me
it.
It
removes
a
possibility
for
for
ourselves
to
have
some.
A
Some
working
base
layer
where
we
can
build
the
agent
on
top
of
we
are.
We
are
kind
of
removing
the
net
beneath
us.
If
we're
going
to
go
in
that
route,.
D
A
A
Yeah
you
have
to
you,
have
to
identify
when
one
command
finished
and
when
the
next
command
started.
So
you
you
would
have
to
print
a
uid
between
each
command
also,
in
order
to
to
to
be
sure
that
you're
not
mixing
the
output
of
different
commands.
That's
super
awkward.
A
D
D
D
There
we
go
sorry,
I'm
back
yeah.
I
turned
my
volume
down
back
yeah.
A
E
Oh,
I
asked
like:
is
it
possible
to
replace
exact
net
with
another
thing.
E
D
D
A
No,
it's
super
easy
to
replace
it
with
some
commands
like
t
and
chmod,
and
that's
that's
everything
we
need.
D
D
I
mean,
maybe
maybe
it
makes
sense
just
to
play
with
it
a
little
bit
see
if
it's
how
hard
it
is.
I
mean
I
don't
know
much
about
the
async
stuff
in
python
I
mean.
Can
you
just
like
write
a
synchronous
function
that
just
does
the
await
and
then
and
then
I
think,
it'll
interoperate,
with
everything
else.
Just
fine.
D
Yeah,
maybe
I
mean
maybe
we
should
just
play
with
play
with
async
stage
a
little
bit
and
just
make
sure
that
it
behaves
the
way
we
think,
because
I'm
it
occurs
to
me.
Also
that,
like
the
way
we're
currently
using
remote
is,
I
think,
a
little
bit
dodgy,
because
like
say
we
do
it,
we
open
a
connection,
we
issue
a
command
and
then
we
like
an
hour,
goes
by,
and
then
we
want
to
issue
another
command
if
something
screwed
up
that
connection.
D
In
the
meantime,
when
you
get
time
to
run
that
second
command
you're
gonna
get
a
failure
right
like
it
really
seems
that
before
you,
if
you're
reusing
a
connection
before
you
do
anything,
you
should
just
do
like
a
a
true
command
on
the
other
side,
just
to
make
sure
that
the
connection
is
healthy
and
if
there's
a
failure,
then
reconnect
and
like
make
sure
you
have
a
health
connection
before
you
actually
do
your
actual
thing
or
else
you'll
get
spurious
failures
on
things
that
you
otherwise
wouldn't
so
it
seems
like
we
could
write
a
pretty
simple
framework
that
just
has
these
persistent
connections
before
we
issue
any
other
and
we
just
make
sure
the
connection's
healthy,
reopen
or
whatever
and
then
run
the
command.
D
A
Yeah,
we
have
to
make
sure
that
the
I
think
ssh
itself
supports
pending.
A
A
A
D
A
Exec
net
is
just
running
remote
python
on
the
remote
host,
got
it
okay
and
etc
is
handling
all
the
ssh
connections
got
it.
Oh
that
is
it
connection,
handling.
Okay,.
C
With
earlier
prestige,
maybe
more
more
to
rgw,
but
thank
you
for
working
on
the
with
our
friedman,
the
archway
m
integration
utology
a
while
back.
Do
you
know
yeah
where
that
ended
up.
D
There
was
a
whole
checklist,
I
think
most
of
the
stuff
that
was
on
that
list
is
in
place
now.
I
think
really
it's.
The
next
step
is
like
going
through
the
test
suite
and
like
converting
things
over
to
use
orchestrator
instead
of
the
rgw
task.
D
D
D
But
I
think
the
big
ones
have
all
been
excuse
me.
I've
all
been
addressed.