►
From YouTube: Ceph Orchestrator Meeting 2021-07-20
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
A
B
A
A
Yeah
mike
is
really
good
at
it,
but
posted
to
youtube.
Maybe
two
weeks
later.
A
A
Yeah,
thank
you
daniel,
it's
even
linked
in
the
other
pad.
If
you
have
a
look
at
the
other
pad,
there
is
a
inline
basic.
There
is
a
recordings
page
anyway.
A
B
A
A
E
Oh,
I
was
just
going
to
say:
there's
also
the
downside
that
this
forces,
the
user
to
understand
the
internals
about
the
orchestrator,
so
some
of
these
variables
are
kind
of
magic.
That's
unfortunate.
A
So
we
have
good
downsides
for
both
approaches,
but
I
guess
that's
it
right.
Is
there
any
other
way
we
can?
We
can
use.
B
I
think
that
maybe
there
is
a
possibility
that
is
to
to
have
separate
places
or
a
standard
templates
and
customized
templates
for
the
user.
So
if
the
user
is
selecting
a
customized
template,
this
is
outside
the
the
official
folder.
That
is
not
going
to
be
the
possibility
to
to
override
the
default
template
that
we
are
deploying
with
with
the
with
the
container
or
with
the
deployment
of
the
fadm.
A
That's
not
the
biggest
issue,
because
a
user
can
always
reset
the
timeover
written
and
let
it
reset
to
the
default
that
we
are
shipping
that
that's
always
a
possibility.
A
It's
just
that
when
doing
a
sf
upgrade
and
we
are
shipping
a
new
template
in
a
new
manager
container
that
new
default
is
never
going
to
end
up
in
the
and
the
deployed
demons,
because
the
user
has
a
config
key
setting
overwriting
the
the
default,
the
templates,
not
the
default
template,
but
the
template.
So
it's
stored
in
the
config
store
in
the
monitor.
So
it's
it's
safe
right.
A
If
we
are
not
forcing
users
to
to
override
the
default
template,
that
would
be
now
that
that
would
be
really
evil,
but
it's
possible
to
store
them
in
convky
key,
but
nevertheless
we
we
have
the
downsides
of
the
upgrade
issue
and
and
the
and
exposing
internals.
A
A
E
E
In
my
mind,
it's
almost
like
there's
more
than
just
like.
We
need
templating
for
some
of
these
things,
but
we
also
need
a
way
for
users
to
define
their
own
custom
configuration
somehow.
E
A
Okay,
we
at
some
point
we
introduced
that
spec
sub
object
in
those
yaml
files
right.
So
if
you
have,
if
you
have
I'm
writing
something
down
in
the
I'm,
writing
something
down
in
the
in
the
other
page
here.
So
if
you
have
a
service.
A
Service
type
ntw
and
you
have
some
additional
properties
of
the
gamma
file
island,
a
spec
object
within
the
yamaha
and
there
we
have.
A
Keep
everything
explicitly
in
in
the
in
in
the
in
the
ncf
edm,
but
we
could
make
it
possible
for
users
to
to
write
additional
arbitrary
files
to
the
spec
file
plus
making
it
possible
to
overwrite
the
template
file
and
then
having
the
possibility
for
users
to
to
first
change
the
template
and
second
use
custom
fields
in
this
in
the
in
spec
files
to
to
override
the
template.
I
don't
know,
does
it
make
sense?
E
E
When
we
do
this-
and
I
think
there
are
some
things
in
the
orchestrator
that
are
hard
to
templatize
so
like
the
http
start
port
for
aj
proxy
things
like
that,
but
maybe
they're
covered
elsewhere.
A
We
could
I
mean
ginger.
2
is
extremely
express
stuff.
A
Okay,
yeah:
we
are
talking
about
that.
F
Oh
yeah,
I
think
I
put
something
in
the
pad.
I
think
I
think
this
is
a
problem,
because
you
can't,
you
can't
see
the
defaults,
you
can't
see
the
current
value.
You
can't
see
how
it
differs.
I
I
think
we
should
just
add,
like
a
whole
new
set
of
cli
commands
that
will
like
list
the
templates,
let
you
fetch
the
current
value
of
the
template.
That's
the
default
value
disk,
the
current
value
with
the
default
and
set
the
defaults
or
set
that
that.
F
Yeah,
it
doesn't
that
we
can,
after
the
upgrade
they
can
then
go
look
at
this
or
it
could
be
that
we
upgrade
when
we
upgrade.
We
we
take
the
disk
and
then
apply
it
to
the
new
one.
I
mean
it's
like.
I
think,
there's
no
good
answer
here.
It's
like
if
you
upgrade
a
debian
package
and
you've
modified
a
configuration
file.
It
always
prompts
you
like.
Do
you
want
to
take
the
new
one,
keep
yours
or
look
at
the
difference
and
maybe
apply
the
disk?
B
I
I
don't
think
that
an
final
user
that
is
modifying
a
template
is
going
to
be
very
happy
if
we
replace
them
again
for
something
new
or
upgrade
it
okay.
So
I
think
that,
if
probably
the
user,
what
he
wants
to
to
have
is
always
the
same
template
and
only
change
in
that
template.
B
If,
if
this
changes,
our
arduino
are
unexecuted
by
by
himself.
F
F
But
probably
the
end
result
is
just
that
after
you
do
the
upgrade
you
should,
as
the
next
step,
go.
Look
at
the
disks
re-examine
your
templates
and
see
if
they're
changed.
D
Not
really
gifting,
all
the
tablets
sounds
really
complicated,
but
it
seems
like
other
than
that.
We
just
have
to
have
like
a
million
different
options
in
the
emails.
F
I
mean
the
upgrade
is
another
issue,
because
it's
like
it's
not
interactive,
so
you
can't
like
prompt
them
for
what
they
want
to
do
with
their
template.
If
it
happens
to
change,
but
you
could
like
log,
something
maybe
or
I
don't
know-
generate
some
report
somewhere,
so
they
can
see
what
the
difference
is
some
of
those
flag
it,
but
I
mean
even
even
independent
of
the
upgrades
like
just
think
having
it
seems
like
at
a
bare
minimum.
F
You
have
to
have
the
ability
to
see
what
the
template
should
be
without
having
to
have
them,
go,
look
and
get
or
like
inside
the
container
image
to
go
copy,
a
file
to
start
with
right,
like
I
should
be
able
to
see
what
the
templates
are
that
can
be
changed.
I
should
see
what
the
current
value
is.
I
should
be
able
to
see
what
the
default
is
like.
That's
I
don't
know
that
feels
like
a
minimum
if
we
like
want
to
invest
any
effort
in
them
being
able
to
customize
these
templates.
F
Otherwise
we
just
say
you're
on
your
own
and
go
read
the
source
code.
Every
time
you
upgrade
to
see.
If
there's
a
problem,
but
this
isn't
supported
you
now,
I
think
we
can
do
better
than
that.
It
feels
like
they
feel
like
comments
like
are,
would
be
pretty
easy
to
implement
a
document
and
if
we
just
say
after
an
upgrade,
go
recheck
your
disks
for
your
templates.
If
you
customize
them,
then
like
that's,
that
would
be
enough.
F
We
can
get
a
notification
of
that
yeah,
but
I
think
for
these
templates
I
think
we
should.
We
should
just
not
make
them
do
that,
but
have
them
stuff
that
have
a
coco
to
set
the
template
separately
right.
A
Would
would
be
great
to
know
if
if
the
info
template
changed
before
user
did
the
custom
template
or
if
the
default
template
got
a
newer
version,
then
a
user
changed
the
set
a
custom
template,
but
you.
A
A
Anyway,
let's
see
we,
we
don't
need
to
decide
that
here
today
we
can
postpone
it
until
we
we,
we
can
move
it
to
the
requests.
A
Discussion
and
then
think
about
it,
maybe
a
day
or
two
okay,
so
another
another
topic
that
we
have
today
is
that
the
manager
is
stuck
for
15
minutes
in
the
surf
loop
daniel.
Do
you
want
to
just
briefly
ex
what
happens
and
why.
C
But
so
basically,
if
you
have
an
offline
host
or
a
host
that
goes
offline,.
C
The
next
time
the
server
comes
around,
it's
gonna
try
to
run
like
gather
facts,
refresh
hosts
and
demons
which
we'll
try
to
run
gather
fact
that
runs
idioms,
gather
facts
and
at
that
point,
cepheum
hasn't
realized
that
the
host
is
offline.
Yet
so
it's
going
to
go
through
the
whole
process
of
it
goes
through
a
bunch
of
functions
and
stuff,
but
it's
eventually
going
to
call
some
remote
functions
that
are
going
to
try
to
either
create
a
connection
to
that
host
or
use
an
existing
one.
C
And
if
the
timing
is
correct,
it'll
have
an
existing
connection
to
that
house.
Even
though
the
host
is
actually
offline
and
then
it's
going
to
try
to
you,
know
ssh
into
the
host
and
run
the
like
gather,
fax
command
and
the
connection
it's
using
isn't
going
to
work
because
the
host
is
offline
and
then
there's
like
a
15
minute
period,
where
it's
trying
to
do
that,
and
it
eventually
will
error
out.
C
But
during
that
15
minutes
the
serve
loop
is
stuck.
It
will
arrow
error
out
after
the
15
minutes
and
mark
the
host
is
offline
after
and
everything
kind
of
goes
back
to
normal.
But
for
15
minutes
it's
done
in
the
tracker,
I
put
a
bunch
of
details
about
what
functions
and
what
lines
are
causing
it
and
what's
happening
and
a
log
of
it
happening.
C
I
don't
really
know
if
there's
a
great
solution.
I
know
melissa
is
working
on
changing
the
whole
ssh
like
backhand
system,
so
maybe
her
basic
ssh
thing
might
be
able
to
resolve
this
or
the
ancient
thing
that
adams
did
is
working
on.
I
don't
know
if
that
would
make
a
big
difference
with
this
or
not
but
yeah.
That's
the
gist
of
the
problem.
A
D
There's
some
way
we
can
just
add,
like
a
timeout
to
this
like
15
minutes,
is
a
really
long
time
to
get
stuck
if
we
could
just
lower
that
I
like
a
lot.
This
is
a
really
rare
timing.
I
feel
like
go
offline
right,
as
this
already
has
a
connection
it's
about
to
do
something
we
could
just
have
that,
maybe
only
like
a
minute.
It
wouldn't
even
really
be
a
big
deal
because
of
how
rare
it
is.
C
There's
not
like
it
well
currently,
there's
no
time
out
that
15,
I'm
not
exactly
sure
where
that
15
minutes
is
coming
from
it's
it's
very
consistent,
15
minutes,
I'm
assuming
it
has
something
to
do
with
the
ssh
or
remote
stuff.
I
don't
know,
I
assume
we
could
probably
add
a
time
up
to
it,
I'm
not
sure,
but.
C
Well,
yeah,
I
mean
if
the
agent
makes
this
problem
go
away.
That's
fine!
Like
I
said
it's
not
the
end
of
the
world.
I
mean
that
the
manager
is
stuck
for
15
minutes,
but
after
that
15
minutes
it
does
resolve
itself
and
mark
the
host
is
offline
and
the
problem
doesn't
continue,
but
there
is
a
small
window
for
50
minutes
where
you're
stuck.
D
I
mean
even
regardless
shouldn't
this
be
super
uncommon
because
you
have
to
like
make
a
connection
like
as
the
host
goes
off
the
line,
and
then
it
has
to
be
like
about
to
do
something
like
hosts.
I
went
offline
super
often,
and
then
it
has
to
be
the
very
specific
timing
when
it
goes
offline
like
unless
you're
purposely
testing
goings
going
offline.
You
wouldn't
see
this
very
often.
C
Yeah
I
mean
I've
been
able
to
reproduce
it
like
very
consistently,
but
obviously
I'm
like
trying
to
like
I'm
manually
putting
hosts
offline
for
that
to
happen.
So
I
don't
know
how
exactly
it
would
behave
in
like
a
real
world
cluster.
I
would
think
I
mean,
based
on
what
I
figured
out.
C
It
seems
like
this
should
happen
almost
every
time,
though,
because
the
only
way
a
host
gets
marked
offline
is,
if
it
it
has,
there's
a
function
in
the
remote
stuff
that
tries
to
create
the
connection
and
if
it
fails
to
do
that,
that's
when
the
host
gets
marked
offline
and
usually
there's
already
an
existing
connection
to
the
host,
and
it's
just
gonna
use
that
and
bypass
the
part
of
the
code
where
it
marks
the
it
could
fail
and
mark
the
host
offline.
D
Yeah
I'll
message
you
later
either
like
I
haven't,
been
I'd,
reproduce
it
personally,
so
I
have
to
look
at
it,
but
if
it's
actually
seems
like
it's
something
that
happens
almost
every
time
it
goes
offline,
we
probably
should
see
if
we
can
maybe
put
a
timeout
in
and
then
it'll,
hopefully
just
go
away.
Let's
replace
this
page
library.
A
All
right
melissa:
do
you
know
if
asking
ssh
supports
the
pcb
keeper
live
setting
of
ssh.
G
Yeah,
I
think
it
does.
I
I
was
reading
something
and
it
does
support
the
keep
alive
function
or
the
tcp
thing,
and
if
it's
related
like
in
my
like
implementation
of
like
the
async
ssh
stuff
for
running
the
commands,
if
the
connection
is
broken,
it
like
returns
an
async,
ssh
error.
And
then
I
just
like
reset
the
connection.
So
I
don't
know
if
that
would
fix
that.