►
From YouTube: 2017-APR-12 :: Ceph Developer Monthly
Description
Monthly developer meeting for the coordination of Ceph project development.
http://wiki.ceph.com/Planning
A
Alright,
here
we
are
another
developer
monthly
for
our
AIPAC
friendly
folks.
Like
stage
mentioned,
it
should
be
a
relatively
short
agenda
unless
we
have
people
showing
up
for
things
that
aren't
on
the
list
already,
but
I
guess
we'll
just
dive
right
into
it
and
knock
out
quickly.
So
you
can
jump
on
that.
First,
one
or.
B
B
So
what
happened
was
well
in
the
beginning,
there
was
calamari
and
it
was
prescriptive
in
terms
of
having
to
install
salt
and
if
it's,
whether
in
a
diamond
with
other
and
I'm
Steph
and
I,
was
a
GUI
on
top
and
they
included
a
REST
API
and
nobody
wanted
to
use
it
outside
of
ink
tanks
because
it
was
too
prescriptive
and
so
there's
basically
their
adoption.
And
then
we
ditched
the
GUI
piece.
We
still
shipped
the
API
sort
of
a
separate
piece
of
part
of
it
had
step
storage.
B
But
there
are
very
few
consumers
that
and
I
know
it's
more
than
zero,
but
I'm
not
sure
how
many
forth
yeah.
So
originally,
when
John
did
the
set
manager
thing.
He
basically
just
took
the
calamari
API
and
just
plopped
it
as
a
manager,
mostly
unchanged
I,
think
fiddled
with.
You
are
a
little
bit
just
a
sort
of
a
demonstration
that,
yes,
the
dad
manager
has
all
the
basic
pieces.
B
So
so
Boris
basically
re-implemented
the
API
using
flask
and
then
again
using
pecan,
because
that
was
less
work
than
packaging
and
then
supporting
forever
after
Django
in
rel,
and
so
we
have
a
pull
request
that
basically
adds
the
rest.
Api,
that's
different,
you
know
didn't
even
you
know
how
different
it
is
from
the
calamari
one
is
it
is
it
similar?
Is
the
distal
reimplementation
in
the
same
finger.
C
D
View
the
worrisome
thing
was
that
there
were
some
things
that
calamari
did
for
long-running
operations
like
sending
a
request
to
increase
the
number
of
PGs
and
then
we'll
be
doing
background
to
find
out
whether
it
actually
succeeded
that
were
kind
of
tricky
to
implement
even
given
django
and
would
be
even
trickier
in
a
simple
web
framework
with
less
asynchronous
nonsense.
Going
on
and
I
know,
John
asked
about
that
under
view.
But
I
don't
understand
where
that
landed.
B
Okay,
yeah
I
know
so
alright,
so
that's
that's
the
as
far
as
people
using
it
we've
told
the
team
making
gooey
stuff
that
they
should
use
something
with
set
manager
right
now
there
you're
like
wrapping
CLI
stuff,
so
they're
not
really
using
it.
So
there's
no,
no
user
or
specific
demand
yet
on
the
right
outside
of
those.
B
Presumably,
we
want
them
using
it's
probably
easier
from
view
that
web,
like
rusty
type,
api's
and
Laredos
directly
but
I'm,
clear
I,
guess
it
depends
on
sort
of
what
the
scoop
is
and
they've
gone
too
far
on
the
open
attic
side
with
suited.
They
are
already
using
something,
but
they
implemented
another.
B
B
So
I'm,
not
an
expert
on
the
again
I,
don't
know
very
much
about
the
new
pecan
room,
but
I
assume
it
includes
all
this
basic
stuff,
right,
stuff
status,
list
of
pools
and
snapshots.
The
racial
code
profiles
OS
these
pts,
like
it
just
exposes
all
this
like
basic
stuff,
I'm,
not
sure
it
exposes
our
biggies,
but
that
would
be
a
pretty
simple
and
good
addition,
presumably
to
that
epi.
B
B
C
The
open
attic
does,
in
addition
to
getting
some
some
status
safe,
and
you
can
also
through
the
open
attic.
Api
can
do
things
like
create
and
delete
tools
and
create
delete,
are
babies
and
things
so
for
each
of
those
endpoints
there's
a
there's,
a
get-well
for
anything.
The
customer
get
and
put
in
a
deleted
can
actually
manipulate
those
things
in
that
restful
way.
The
the
salt
stuff
that
wouldn't
go
anywhere
near
yeah.
C
B
Okay,
so
there's
that
data
point
and
then
the
last
thing
is
so
John
followed
up
on
the
pull
request
with
sort
of
asking.
Why
wasn't
this
discussed
on
death
develop
before
the
fixing?
Was
that
can't
really
blame
for
us
because
he
actually
asked
him
to
open
the
PR,
so
they
comment
on
it,
but
I
think
that
his
larger
point
was
that
we
haven't
really
discussed
what
API
we
want
and
John
is
questioning
whether
we
actually
do
want
a
REST
API,
because
we
don't
really
have
any
users.
Yet.
B
Obviously
we
don't
have
one
where
people
specifically
demand
it,
and
even
though
people
want
a
management
API,
it
wasn't
immediately
clear
to
John
that
it
should
be
a
restful
one.
It
could
be.
You
know,
XML,
RPC
or
I,
don't
know
whatever,
whatever
people
use
JRPG,
where
people
in
two
days
I'm,
pretty
ignorant
of
this
whole
field
of
people
building
wrestled
things
with
Python
frameworks.
Whatever
is
not
really
my
thing,
so
I
don't
have
any
strong
opinion,
except
that
there
was
a
pull
request
that
just
came
in
to
greatest
gateway.
B
Like
two
weeks
ago,
I
was
basically
trying
to
add
a
management
API
endpoint
to
Rios
gateway,
because
they're
like
it's,
really
can
be
easy
to
write,
automation
and
management
tools.
If
there's
a
restful
endpoint
and
we
should
just
put
it
in
the
gateway,
because
it's
also
already
doing
you
know
HTTP
stuff
and
clearly
it
should
not
be
in
rid
of
gateway.
It's
the
game
that
it
should
be
a
manager,
so
I
think
we
need
like
it's.
My
feeling
is.
We
need
something
Boris's
for
requests.
C
I'm
I
mean
a
slightly
odd
middleman
position
at
the
moment,
because
I'm
in
sufficiently
familiar
with
the
open
edit
code
base
right
now,
but
so
the
the
reason
that,
just
for
a
little
bit
of
background,
open
attic,
did
a
REST
API
internal
to
open
attic
for
talking
to
Saif
at
the
time.
Because
manager
didn't
exist
yet
and-
and
there
was
there
were,
was
at
theum.
C
There
were.
There
were
things
that
were
annoying
somehow
about
the
the
calamari
api
now
and
that
the
open
attic
the
open
had
excess
API
is
a
Django
thing
at
the
moment,
because
the
whole
apps,
Django
and
I
had
actually
thought
to
just
sort
of
pick
that
up
and
drop
it
into
manager
and
see
what
happened.
I
haven't
actually
done
that
yet,
but
just
as
an
experiment
and
I
had
a
little
bit
of
a
little
bit
of
email
back
and
forth
with
a
couple
of
the
open
attic
guys
last
night,
my
time
and.
C
Some
of
the
teal
they
are
they
at
some
point
need
to
move
their
REST
API
stuff
to
submit
need
to
be
using
a
REST,
API
incest,
manager
right
because,
for
example,
they
want
to
get
PG
and
OSC
dumps
and
doing
it
the
way
they're
doing
at
the
moment.
It
will
time
out
on
big
clusters
somehow
and
doing
doing
these
sorts
of
things
through
self
manager
through
a
REST
API
there
is,
is
going
to
be
a
good
fit.
C
C
C
C
B
C
C
I'm
not
I,
mean
I'm
trying
to
do
work
on
manager
at
the
moment,
but,
like
I,
said
I'm
not
deeply
familiar
with
the
open
attic
bits
at
the
moment
and
I'm
not
sure
how
I'm
not
sure
if
anybody
else
in
the
arithmetic
team
has
an
enormous
amount
of
time
before
the
Luminess
release
to
do
stuff
on
this
so
again,
without
actually
trying
to
speak
for
anybody
else,
and
so
one
question
that
one
question
would
be
is
is
is
do
we
want
this
set
this
API
to
go
in,
as
is
for
luminous
and
then
fiddle
with
it
more
later
or
do
we
want
to
hang
back.
B
Probably
most
of
these
are
going
to
be
pretty
easy
to
just
map
onto
what,
when
you
in
add
you
can
add
the
RVD
stuff
and
back
port
it
if
we
don't
manage
to
get
it
in
before
remembers
release,
although
probably
I'm
guessing
it
isn't
all
that
much
work
to
do
it,
and
so
she
a
killer
barbecue
since
there's
no
additional
yeah
I,
think
I'm
gliding,
Boulevard
II
stuff
into
manager
tree
really
something.
I
guess
I'm
not
yesterday,
but
on
them
I'm
curious.
What
the
other
I
guess,
I
guess
a!
Are
there
other
concerns
and
be.
C
F
C
C
My
concern
is,
if
this
is
a
port
of
the
calamari
API,
just
to
make
it
simpler
without
further
thought
into
what
pieces
various
users
might
need,
and
what
and
if
this
is
the
API,
that
we
want
to
maintain
going
forwards.
If
this
gets
a
sort
of
solidified
and
declared
as
the
thing-
and
we
can't
you
know
like
right
now
without
enough
users
to
know
whether
it's
the
right
thing,
then
I'd
only
get
painted
into
a
corner
yeah,
but.
B
D
B
D
G
B
E
B
Original
cell
amore
it
may
or
may
not
have
been
sort
of
ordered
over
to
this.
Maybe
maybe
just
got
dropped.
I
think
maybe
does
that
drop,
but
so
what's
there,
what
should
be
there?
I
mean
I,
don't
have
to
implement
that
immediately.
But
if
there
is
something
there
then
we
should
like
make
sure
it's
totally
wrong.
B
C
B
C
B
B
B
Next,
all
that
I
have
here
I
guess.
The
last
thing
is
that
there's
the
Deaf
rest
api
endpoint,
that's
basically
just
a
pass-through
for
the
deal,
I,
KPI
and
narshall
vessel
in
HTTP
form
and
that
there
is
something
similar
in
the
calamari
API
just
with
a
different
end
point
that
also
does
passive
commands
do
probably
in
this.
Whatever
this
implementation
will
fix,
we
should
still
have
a
similar
passer,
where
we
can
do
send,
monitor,
commands
a
little
bit
proxy
straight
to
the
reminder.
B
So
then
we
can
take
the
existing,
take
the
existing
tests
and
just
change
the
endpoint
and
run
them
to
the
manager,
and
then
we
can
just
remove
the
separate
API
from
encode
tree.
Finally,
that
it
would
be
nice
if
that
could
get
wrapped
up
before
Illumina.
So
we
have
a
little
bit
less
proliferation
of
stuff.
B
E
B
Okay,
that
guy
with
that,
so
thanks
thanks
Tim,
looking
forward
to
hearing
what
you
find
and
then
the
other
yep.
The
only
other
thing
I
have
on
here
is
a
set
config
option.
Annotation,
so
I've
been
sort
of
I've
been
putting
together
a
list
of
like
stuff
that
I
haven't
able
to
think
of
that
users
run
into
that's
complicated
or
hard
to
configure
or
just
sort
of
mix,
F
harder
to
use
or
administer
that
none
could
be
not
as
hard
and
make.
B
This
is
some
more
self
managing
any
do
to
use
and
for
the
one
of
the
things
that
we
get
from
users
and
from
customers
and
villages
that
that's
as
hard
and
like
come
up
with
all
the
things
that
configure
to
mitigate
that.
So
there
are
several
things
going
into
luminous
that
are
going
to
mitigate
a
bunch
of
like
things
that
people
run
into
in
the
past.
B
Like
crush
rules
for
multiple
device
classes,
going
to
be
sort
of
done
out
of
the
box,
you
want
to
manually
bid
at
a
crush
maps,
disk
replacement,
we're
documenting
clearly
and
succinctly
make
that
as
more
streamlined
process,
there's
going
to
be
a
manager
endpoint,
that's
just
like
a
set
status
like
minimal
GUI.
That
just
shows
you
know.
Health,
okay
and
log
messages
that
sort
of
thing
and
balancing
a
crush
map
stuff
like
that.
B
But
one
of
the
other
one
of
the
recurring
complaints
is
that
the
config
options
there
are
a
million
of
them
and
the
league
knows
what
they
mean
in
the
document
and
it's
never
clear
which
ones
you
should
be
touching
in
which
ones
you
should.
When
we
talk
about
this
like
years
ago
about
like
annotating,
you
can
pick
options
with
a
description
and
having
a
command
that
will
like
dump
the
convict
options
with
the
description.
B
So
you
can
tell
what
the
hell's
going
on
so
I
just
want
to
resurrect
that,
and
the
kind
of
thing
that,
like
bang
out
on
a
airplane,
once
you
go,
write
descriptions
for
every
config
option
where
ever
once
you
show
to
figure
out
what
that
what
the
scheme
would
be.
So
I
wanted
to
talk
about
that
and
get
input
on
what
like
what
what
properties
are
important
for
config
options,
which
ones
make
sense
and
how
we
should
sort
of
boxy
them,
so
the
I'll
taste
you
could
pad
in
the
chat.
B
B
So
if
you
look
the
currently
the
config
options
have
a
few
properties
there's
the
name
of
the
option.
Obviously
there's
a
type
that's
like
integer
string,
signed,
unsigned,
float
whatever
and
optionally.
There's
a
validator
function
that
you
can
have
a
callback
that
make
sure
that's
about
the
big
option.
B
There's
whether
or
not
there's
an
observer
indicates
whether
they're,
safe
and
there's
a
default
value,
of
course,
but
mostly
that's
it,
and
the
description
for
the
options
are
comments
in
that
in
the
source
file,
which
is
not
very
useful,
so
I
want
to
expand
that
macro.
So
that
includes
all
this
other
stuff,
like
a
description
like
a
short
one-line
description
of
what
the
option
is
and
what
it
does,
and
mostly
maybe
a
longer
string.
That's
like
a
longer
like
paragraph
length,
description
that
there
are
sort
of
more
information
about
that
that
should
be
captured.
B
So
that
I
think
that's
pretty
uncontroversial.
Probably
the
new
idea
that
I
had
that
we
hadn't
talked
about
before
was
the
idea
of
having
a
memory.
We
did
having
a
level
associated
with
the
option
that
so
options
that
users
are
expected
to
adjust
would
be
sort
of
a
low
level,
and
then
things
that
are
here
probably
shouldn't
change
and
be
a
higher
level
and
things
that
are
like
only
developers
are.
B
B
So
that
was
and
then,
along
with
that,
you
could
have
a
config
option.
That
is
like
what
level
am
I
allowed
to
change,
and
so
by
default
you
would
only
be
able
to
change
the
you
know,
like
goodness,
at
the
mid-level
commands
and
if
you
try
to
override
ones
that
were
above,
that
level
would
just
ignore
it
or
issue
a
warning,
or
something
like
that
unless
you,
until
you
change
it
to
like
expert
mode
or
something
like
that,
so
that
you
can
adjust
the
higher
level
options.
B
H
B
A
little
bitty
enough
I
was
trying
to
figure
out
how
to
satisfy
them
books,
okay,
so
associated
with.
That,
though,
is
when
I
did
the
similar
thing
with
the
perf
Kanner's
I
sort
of
defined
it
I
did
predefined
values,
but
I
put
them
on
a
scale
of
one
to
ten,
so
you
could
sort
of
have
stuff
that
was
in
between
if
you
needed
for
more
granularity,
and
they
make
sense
to
do
anything
about,
or
should
we
just
like
keep
it.
Thank.
B
I
Page
oh-oh-oh
be
the
double
red,
because
here
it's
seductive
and
dangerous,
I
think
in
a
certain
way,
because
right
now,
if
you
look
in
config
apps,
we've
got
something
like
ninety
seven
groups
or
options
and
I
think
Tony
blue
FS
options.
That's
just
a
small
subset
of
everything
that
we
have.
It
I
think
when
you
use
create
levels
like
this.
It
account
rolls
you
into
a
false
sense
of
security
by
hiding
a
bunch
of
stuff
from
the
user
that
they
probably
you're
right.
I
They'll
never
need
to
touch,
but
it
it's
a
lot
of
additional
things
that
theoretically
a
developer
multiple
developers
are
thinking
about.
But
in
reality
you
know
if
anyone
actually
looking
at
these
is
anyone
actually
making
sure
they're
right,
maybe
someone
is
maybe
I
am
maybe
bending
windows,
maybe
other
developers
are
but
I
worried
that
it's
kind
of
a
way
to
shove,
everything
under
the
bed
sort
of
does
that.
I
A
F
B
B
You
know
like
replacing
the
million
throttling
things
with
the
recovery
things
that
control
recovery
rate
and
said
with
a
single
option
either
says
I
wanted
about
this
percentage
as
my
throughput
or
whatever
to
recovery,
and
then
figure
out
what
to
do
based
on
that
right.
That
would
be
a
sort
of
a
pretty
good
impose
yeah.
I
Kind
of
the
same
thing,
with
all
the
memory
that
we
allocate
different
states
right,
I
mean
we've
got
now.
We've
got
like
an
internal
rock.
We've
got
two
different
internal
rocks,
DB
caches,
there's
a
compressor
and
an
uncompressed
one.
We
have
oh,
no
Andy
store.
We
have
you
know
other
things
all
over
the
other
place.
We
have
never
be
bored
of
blue
FS
cash
on
top
of
all
the
other
caches.
It's
just
you
know,
there's
we
it's
a
lot
of
effort
right,
but
we
could
probably
make
such
smart
about
saying.
B
Now
is
take
the
like
completely
opaque,
all
of
messiness
and
and
put
some
guardrails
or
in
its.
They
don't
wander
out
for
these
because,
like
story
story,
some
customers
who
are
like
looking
at
all
the
config
options
being
totally
overwhelmed,
having
no
idea
what
any
of
them
does
and
just
like
spending
them
and
then
seeing
what
happened
to
the
cluster
and
then
changing
another
one
and
say
I
have
requests
here
and
you're,
like
that's,
like
that's
worse
way,
to
spend
your
time
basically
yeah.
So.
B
So
for
the
things
that
are
under
dev,
it's
going
to
be
like
failure
injections,
and
it's
going
to
be
like
things
that
inject
delays,
so
that
test
can
see
them
or
just
like
weird
things
that
are
designed
to
make
the
code
behave
in
a
weird
way,
so
that
you
can
make
sure
you're
tested,
particular
I.
Don't
know
if
that
will
fail
for
this
team,
I
mean
if
there
were
a
test
category
I'm,
not
sure
that
could
be
that
many
things
that
going
bad
but
not
test,
I,
guess:
okay,
yeah.
B
B
H
H
B
So
that's
the
next
thing
in
the
pad.
The
the
next
one
on
this
list
is
I.
Guess,
there's
no
category
but
there's
an
effects
string.
Any
idea
there
is
that
you
would
list.
You
just
have
a
list
of
the
things
that
it
affects,
so
the
option
affects
them
on
Ernie
OC,
only
you
would
list
those
or
if
it
affects
like
the
common
code
and
everything
you'd
say
that
or
yeah.
So
you
could.
It
would
be
a
list
and
imagines
a
list.
B
B
B
B
B
Yeah
yeah
yeah,
so
that's
my
thought.
So
if
you
each
option
would
have
a
list
of
the
things
that
effects
and
then,
when
an
actual
demon
is
initializing
by
calling
common
in
it
see
what
it
was
right.
Its
list
of
the
things
that
it
is
so
like
LaBarbera
d
would
list,
live,
RVD
and
labret
dos,
probably
and
common
of
its
its
list
of
things,
and
then
it
can
take
to
someone
know
that
that's
the
fit
overlapping
said
it
can
take
options,
other
ones
that
M
that
it
needs
to
pay
attention
to.
H
B
Maybe
I
think
ideally
there'd
be
like
there's
this
common
pattern
of
people
they're
a
bunch
of
options
that
are
prefixed
with
OSD
that
control
how
pools
are
created
like
the
default
crush
rules
is
like.
Oh,
the
pool
default,
crush
rule
or
something
like
that
and
it's
used
by
the
monitor
right.
But
it's
fine,
but
it's
confusing
because
names
OC
it
so
people
so
it'd
be
nice.
It's
there.
B
The
warning,
if
you
like,
set
the
wrong
to
take
option
on
the
right
thing,
but
I'm
not
sure
that
that
really
works,
because
the
way
that
the
config
options
are
applied
is
hierarchical.
So
you
have
to
get
all
clever
like
it's.
A
optional
set
a
specific
OSD
or
under
the
osc
category,
but
does
not
pick
those
two
I
don't
know.
So
maybe
we
just
ignore
that
and
just
because
we
could
populate
those
fields
using
just
a
documentation
now
but
I
in
the
interest
of
making
this
something
that
we
can
actually
accomplish
and
add
value.
J
A
D
D
K
B
I
So
one
thing
I
noticed
on
the
performance
side
is
that
we've
got
a
lot
of
users
that
have
a
problem
right.
They
think
things
aren't
going
as
fast
as
it
should
and
they've
seen.
Someone
say:
oh
just
increase
your
number
of
Oh
with
the
objects,
and
that
will
give
you
a
performance
boost
or
something
you
know
and
there's
this
kind
of
tribal
knowledge
that
goes
around.
I
That
doesn't
necessarily
do
anything
for
them,
but
the
the
word
of
their
problem
might
be.
They
bought
an
SSD
that
doesn't
do,
desync,
writes
fast
or
you
know,
whatever
it
could
be
millions
of
different
things.
So
how
do
we?
Even
if
we
do
all
this?
Are
we
still
going
to
have
a
ton
of
users
that
are
just
kind
of
wildly
stabbing
at
changing
different
parameters?
To
maybe
explain
to
do
things
for
them,
I
mean.
B
Probably,
but
there's
there's
no
way
to
like
control
the
tribal
knowledge
right,
I
think
the
best
we
can
do
is
like
a
try
to
get
them
to
share
it
in
a
forum
where
there's
some
peer
review,
I
guess
consider.
The
the
other
thing
I
would
like
to
do
is
try
to
get
them
to
the
tribal
knowledge
and
build
into
the
system
so
that,
if
you
have
sse,
then
COC
automatically
increases
its
default
up
thread,
size
right
and
just
sort
of
unrelated
to
this
specifically,
but.
B
Does
that
actually
accomplish?
Anything
I
think
is
I.
Guess
why
I'm
asking
given
the
conversations
I
heard
today,
I
think
it
does
okay,
well,
certainly
the
documentation.
That's
like
right
now,
we're
manually,
maintaining,
even
if
all
we
did
was
generate
rst
Doc's
from
us,
and
that
was
like
it.
It
would
I
think
would
still
be
worth
it
because
they
would
like
manually
maintaining
that
in
a
separate
rfp
file.
That
nobody's
remembers,
updates
and
updates.
Oh
okay,
so
all
right
so
back
to
the
affecting.
B
Maybe
just
we
just
simple:
let's
apply
for
something
like
tags
and
we
can
sort
of
how
exactly
exact
metrics,
so
that
can
be
TBD,
but
for
now
we
just
have
a
simple
set
of
like
which
demon
to
detect
or
whether
it's
common,
because
that's
just
having
the
tags
of
sort
of
a
formal
documentation
where
you're
like.
Oh,
this
is
an
OC
option
or
whatever
and.
B
Then
it's
a
couple
other
two
other
random
things
here.
One
is
that
if
it's
a
numeric
value,
we
could
define
an
event
in
a
max.
They
don't
set
like
you
know,
negative
one
on
a
time
value
or
something,
and
for
for
string
options
are
a
bunch
of
string
options
where
it's
actually
a
really
Nina,
where
there,
like
a
certain
specific
number
of
strings
that
are
recognized
and
others
are
not
plowed
that
it's
just
an
option.
You
can
set
a
data
stream.
B
D
B
B
Yeah
they're
often
like
pairs
of
options
or
set
the
three
or
whatever
they
go
together.
So
that
would
be
good,
okay,
okay,
so
so
the
further
config
levels,
if
either
expert
InBev,
should
use
either
the
right
primers
that,
like
Edmund
I,
don't
know
I
would.
B
B
Thanks
well,
okay!
So
regardless,
then
we
needed.
E
E
B
I
B
G
I
E
B
Prevent
people
from
changing
and
we
want
them
to
know
that
they're
a
dangerous
territory
so,
like
you
know
what
SSL
or
what
your
the
rgw
DNS
name
for
your
wildcard
DNS,
like
that's
a
basic
option.
That's
something
that
you're,
normally
a
normal
user
is
going
to
set
and
and
if
you
can
think
about
it,
if
somebody
has
to
go
in
and
chain
an
advanced
option,
that's
sort
of
like
a
failing
on
our
part
that
we
can
do
it
automatically
or
something
right
and.
B
B
E
D
B
B
Well,
her
the
other.
The
other
part
of
this
was
to
have
a
config
option
that
basically
controls
will
prevent
you
from
changing
things
above
a
particular
level,
so
by
default.
That
would
be
set
that
basic
I
guess
so,
like
max
config
level.
Equals
basic
means
that
if
you
go
and
set
an
advanced
option,
it'll
just
issue
a
warning
and
ignore
it.
I
can't
tell
that's
a
good
idea,
it's
sort
of
like
in
like
the
Cisco
IOS,
whatever
to
help
out.
J
B
J
B
At
the
guardrail,
not
you
can't
prevent
malicious
things.
Yeah
I
mean
in
order
to
in
order
set
these
options,
you
have
a
set
and
that
config
level
is
advanced
and
then
set
the
advanced
option
like
it's
not
like
you're,
preventing
the
from
doing
it.
You're
just
making
sure
they've
like
made
it
explicit
another
month,
they're
going
and
they're
going
out
of
the
but.
I
B
I
B
F
B
Okay,
cool
one,
so
the
last
piece
of
this
is
just
a
bunch
of
new
a
stock
commands
that
that
show
the
stuff
so,
where
you
have
config
show
getting
set
I
think
there's
a
pull
request
that
doesn't
config
dish,
but
it
hasn't
merged
that
maybe
rebase
and
cleaned
up.
But
abase
would
just
only
show
the
config
options
that
differ
from
the
default
and
are.
B
B
B
B
B
J
B
So
I
thinking
that,
if
you
should
take
help
with
no
argument,
it
would
show
whatever
is
below
your
current
mode
threshold,
MSU
added
increment,
then
it
would,
you
could
say,
can
take
help
advance
and
it
would
include
everything
up
to
advanced
or
config
help.
Devon
it
week
lose
everything
up
to
step.
Right.
I
got
some
TV
yeah,
Mobley
close
and
star
dev
or
whatever.
B
Okay:
okay,
so
though
this
is
all
often
like
min
and
Max,
and
you
know
and
stuff
like
that,
I
think
yes,
see
really
I
think
that
first
pass
is
just
so
like
documents
them
and
then
they'll
be
some
annoying
grip
to
generate
RT
and
then
wire
that
into
the
like.
The
set
Docs
generator
can
construct,
and
hopefully
talk
can
be
open.
Doing
that.
B
B
J
B
B
Up
probably
use
our
SP
to
put
this
to
make
the
initial
path
of
this,
but
the
goal
is
to
have
it
documented
in
the
header
file
and
it
config
offs
header
file.
It
would
be
right
there
with
the
configure
option,
so
it
would
be
impossible
to
add
a
config
option
without
writing
a
documentation,
basically
right.
B
J
B
No,
that
is
a
very
good
idea,
so
we
could
add
a
we
could
tag
to
get
a
tag.
Deprecated
or
gonna
have
a
pool.
I
B
F
B
B
I
guess
safe
would
fall
into
that
category
observed
community
is
explicitly.
We
have
this
sort
of
annoying
situation
right
now,
where
everything
is
sort
of
unclear,
whether
you
can
change
the
line
or
not.
There's
some
that
we
know
you
can
change
and
there
and
then
there's
everything
else
so
requiring
restart
would
mean
that
we
know
which
ones
you
can't
change,
but
it's
in
there.
So
all
the
in-between
and
eventually,
hopefully,
though
it
would
be,
it
only
two
sets.
But
right
now
we
have
three.
B
Maintain
any
flags
I
mean
we're
giving
them
in
sync
with
real
rope.
Yeah
well,
like
I
mean
the
system
can
could
worn
so
when,
when
everything
starts
up
and
it
registers
all
the
config
observers,
if
you
observe
an
option
that
requires
restart
like
I
said,
that's
fine,
because
sometimes
you
so
yeah
I
mean
it.
Could
it
could
like
assert
if
you
do
that,
if
you
want
to
put
their.
B
J
B
J
K
D
B
B
B
D
D
And
yeah,
like
I,
said
that
every
bases
on
master
that's
running
today
and
he's
chosen
to
use
the
monitor,
config
II
options
for
her
configuration,
so
I've
done
that
for
the
two,
the
two
options:
I've
added
for
this
important
was
matter
which
probably
will
want
to
soon
have
a
discussion
about
what
that
namespace
should
look
like
right
now.