►
From YouTube: JupyterLab Dev Meeting, November 18, 2016
Description
Meeting of the JupyterLab development team, November 18, 2016.
Meeting Notes: https://jupyter.hackpad.com/JupyterLabNotebook-Weekly-Meetings-UUJ3gIQ3iBS
B
And
trying
to
aggressively
attack
the
beta
issues
milestone
on
Jupiter
lab
closed
quite
a
few
this
week,
so
good
progress,
I'm
working
closely
with
the
our
brains.
Folks,
on
the
abstract
editor,
we
had
some
in-person
discussion,
Jason
Ian
myself
last
week
and
then
Alex
and
Anton
from
our
brain
are
helping
move
that
work
forward,
so
I
think
we're
getting
close
to
a
good
solution
and
work
everybody
and
also
been
working
with
darian
this
week
on
the
state
saving
refresh
data
basement
that
he
can
talk
about
further.
C
Okay,
I
guess,
although
yeah
so
like
Steve,
said,
there's
a
new
concept,
top-level
plug-in
in
Jupiter
lab
called
state
DB,
and
it's
it's
both
an
interface
and
an
implementation.
The
interface
is
for
just
a
generic
database
that
takes
just
a
generic
key
key
value
store
and
it
can
be
pulled
for
the
max
length
of
what
it's
going
to
store
the
the
idea
is
that
you
can
give
it
any
JSON
value,
serializes
it
and
stores
it.
C
So
the
max
length
actually
is
the
length
of
the
string
which
can
be
confusing
because
javascript
strings
aren't
predictable
and
the
size
they
might
actually
be
when
they
get
stored,
and
you
know
that
there
might
be
some
confusion,
but
chances
are
pretty
good.
It
won't
come
up
because
we're
not
going
to
store
massive
massive
things.
The
current
implementation
as
its
that's
its
back
end.
It
just
uses
local
storage,
but
of
course
you
could
imagine
a
server-based
or
whatever
else,
and
so
all
of
the
all
the
calls
to
it
return
promises
in
case.
C
We
need
to
do
an
async
version
and
right
now,
because
we
don't
have
a
way
of
restoring
layout.
Yet
all
we're
doing
is
we're
just
restoring
the
existence
of
components
that
were
there
before
you
hit
refresh
so
I
might
have
had
for
notebooks
open
and
they
might
each
have
had
their
own
separate,
a
mini
window
in
the
doc
panel,
when
I
refresh
the
page
I'll
get
four
tabs
in
the
main
doc
panel
with
all
four
notebooks,
so
I
will
have
lost
the
layout,
but
I
will
have
retained
the
notebook
right
now.
C
It
supports
just
notebooks
consoles
and
terminals,
but
I
have
an
open
PR
to
do
editors
and
to
do
CSV,
widgets
and
just
working
through
some
of
the
things
that
can
go
wrong.
So
if
you
restart
your
server,
then
it'll
try
to
recreate
a
console
with
a
session
that
no
longer
exists,
so
I
needed
to
fail
gracefully.
C
If
you
delete
a
file
and
you
delete
a
CSV,
for
example,
I
trying
to
figure
out
what's
a
good
way
for
it
to
try
to
load
and
then
feel
how
gracefully,
but
basically,
the
the
minimum
experience
of
refreshing
the
page
and
not
losing
your
stuff
might
require
layout
to
be
retained
as
well,
but
at
least
we're
one
step
closer.
Now.
Yeah,
so
that's
that.
C
It
doesn't
have
to
be
the
idea.
It
depends
on
what
sorts
of
things
we
want
to
store
where
so,
for
example,
we
might
we
might
introduce
a
settings
plugin
that
for
the
first
implementation
uses
this
and
just
sits
in
your
browser,
local
storage,
and
then
we
upgrade
it
to
use
the
back
end.
But
we
might
never
want
the
back
end
to
hold
your
session
state
when
you
refresh
the
page.
So
we
might
pick
some
things
that
belong
server-side
and
some
things
that
blown
client-side.
C
C
It
might
even
be
the
case
that
we
come
up
with
a
like
a
flag
that
you
send
in
when
you
say,
save
like
save
local
or
safe
server
side,
or
might
be
the
case
that
we
come
up
with
a
separate
plugin.
That's
persistent
state
DB
versus
like
non
persistency,
TV
I'm,
not
sure
yet,
but
but
currently
it's
just
using
your
browser,
local
storage
and
really
the
use
case.
We
have
in
mind.
Is
you
just
hit
refresh
and
we
don't
want
you
to
be
annoying
anymore
right.
C
Yeah,
so
so
right
now,
it's
meant
to
be
used
as
a
plug-in
with
one
static
instance,
but
we
can
imagine
another
plug-in
that
actually
adheres
to
the
same
exact
interface
and
it's.
It
is
also
a
static
instance,
but
it's
an
instance
that
talks
to
a
server-based
back
end
so
yeah.
So
the
the
main
idea
is
to
get
that
interface
right
and
to
get
everyone
using
using
an
abstraction
over
whatever
storage
were
actually
we're.
Actually
gonna
put
into
the
implementation
I.
A
Think
that
that
sounds
like
a
fantastic
approach,
cool
anything
else
from
your
side
during
and
that's
it
for
me,
okay,
I
guess,
I'm
the
next
person
on
the
half
pad
so
I've
mostly
been
working
on
some
design
related
things,
Cameron's
been
doing
a
lot
of
homework
and
investigation
on
font,
icons
and
or
icon
fonts,
and
also
SVG
based
icons
and
there's
a
very
nice
blog
post
that
actually
let
me
post
on
the
hack
pad
so
folks
can
beat
it
recently
at
the
in
2016
earlier
in
February,
github
transitioned
over
to
delivering
all
of
their
icons
using
SVG,
and
they
have
a
fantastic
blog
post
about
that,
and
it
is
now
the
heck
up.
A
The
back
pad
is
behaving
oddly.
There
we
go,
and
so
Cameron
has
started
a
pull
request.
Trying
that
out,
one
thing
that
immediately
showed
up
is
that
the
the
development
and
design
workflow
becomes
a
thousand
times
easier
in
that
any
of
our
designers
can
then
just
open
SVG
files
and
edit
them
and
those
are
stored
in
the
repo
and
used
as
is
and
so
there's
no.
It
looks
like
there's
no
rendering
artifacts
with
SVG
based
icons.
A
A
My
guess
is
that
it
will
be
fine
in
that
github
right
now
has
likely
many
more
icons
on
their
typical
page
than
we
will
have
and
I
don't
think.
That's
been
an
issue
for
any
of
us.
I've
never
heard
people
complaining
about
that.
The
performance
issues
on
github
but
will
submit
a
camera,
will
submit
a
pull
request
and
look
at
that
and
then.
B
A
C
Go
ahead
all
right
and
the
the
sum
of
the
SPG's
we
have
are
embedding
pngs
inside
their
base.
64
encoded
PNG's
inside
the
SVG
soft
that
and
I
said
that
SVG
is
great.
Just
make
sure
you
have
full
SVG
all
the
way
down,
so
we're
actually
doing
it
right,
but
ya,
know
I.
Think
this
approach
sounds
great.
Otherwise,
ya.
A
Know
we
can
definitely
work
with
Cameron
to
make
sure
that
we've
got
sort
of
pure
and
simple
svgs
that
load
quickly.
It's
likely
that
I
mean
what
he's
trying
right
now
just
to
make
it
simple
is
just
copying
over
the
needed
icons
from
material
design,
but
not
the
whole
material
design,
repo,
that's
huge,
just
the
icons
we
need
in
just
in
the
SVG
format.
A
A
Don't
have
an
icon
to
denote
Jupiter
notebook
and
that's
something
we
like
we
want
to
have
and
also
I
think
Cameron
wants
to
start
to
come
up
with
sort
of
general
principles
for
icons,
icon
design
in
Jupiter
lab,
in
that
a
lot
of
other
people
who
are
starting
to
write
plugins
are
going
to
end
up
having
icons
as
well
be
nice
to
be
able
to
tell
them.
You
know
here
the
aspect
ratios
of
icons
you
should
designed
for
here's.
A
The
padding
around
the
icon,
you
should
design
for
font
sizes
or
icon
sizes-
things
like
that,
but
we'll
keep
everyone
posted.
He
was
making
good
good
progress
on
that
the
others
I
have
been
doing
some
work
sort
of
trying
to
assess
how
we
get
from
where
we
are
today
to
the
beta
and
the
one
dodo
released
from
a
interaction
and
visual
design
perspective,
and
so
I've
been
spending
time.
Studying
this
I
think
some
initial
thoughts.
A
One
is
that
one
thing
we
found
when
we
release
the
original
notebook
is
that
the
time
between
us
being
feature
complete
in
the
original
notebook
at
the
time
that's
releasing
because
of
sort
of
design
refinements
and
mostly
not
in
a
visual
design,
but
a
lot
of
just
interaction.
Design
UX
was
quite
long
and
I
just
want
to
flag
for
everyone
that,
like
we
need
to
allow
some
amount
of
time
from
being
sort
of
feature
complete
to
being
release
ready
and
debate.
It
can
be
part
of
that.
A
A
We
want
ourselves
to
become
frustrated
with
the
design
before
our
users
do
I
think
so
I
don't
know
if
people
are
willing
to
do
that.
I'm
trying
to
do
that
so
over
the
weekend.
It
basically
just
worked
in
jupiter
lab
I'd
love
to
before
I
talk
more
I'd
love
to
hear
comments
on
that
idea,
having
a
seat
refresh.
C
That
yeah,
that
does
help
that
I
just
laughed
out
about
with
the
mute
on
so
I
figured
I
should
I
mean
Lee,
so
it
sounds.
It
sounds
like
a
really
good
way
to
intimately
become
familiar
with
things
that
might
bug
people,
but
it's
also
the
case
that
the
kind
of
development
that
goes
into
building
the
application
is
really
devoted
different
than
the
development
that
goes
into
creating
notebooks.
C
A
And
I
I
think
the
main
parts
that
we'd
be
able
to
test
this
way
would
be
sort
of
the
the
overall
application,
shell,
the
menus,
the
command
palette,
the
file
browser
and
the
terminal
and
text-editor.
Obviously,
you
know
work
we're
not
writing
notebooks
when
we
develop
Jupiter
lab
or
writing
in
typescript
and
writing
typescript.
C
Yeah,
it's
good
idea
and
there's
definitely
it's
a
it's
a
heterogeneous
set
of
files
that
we
work
on
and
some
of
them
are
substantially
easier
to
to
do
outside
of
an
IDE
than
others,
so
yeah
for
sure,
yep.
A
A
It's
like
you
honestly,
a
lot
of
the
things
he
ends
up.
Talking
about.
You
could
imagine.
Oh
my
gosh
yeah
this.
This
culture
would
be
incredibly
toxic
and
difficult
for
non-programmers,
but
also
for
minorities,
underrepresented
groups
and
a
lot
of
the
lot
of
the
things
that
are
in
that
book.
I
hear
a
lot
from
the
non-programmers
in
our
project,
the
designers
on
ax
and
another
some
of
the
business
students
we've
had
where
and
so
I
don't
know
it's
again.
A
I
I
identify
myself
as
a
programmer
and
I
don't
get
to
program
as
much
as
a
lot
of
you.
But
it's
it's
a
really
challenging
book.
I,
don't
know
if
folks
are
interested
in
looking
at
it,
but
it's
really
really
thoughtful
about
sort
of
how
we
build
software
and
how
that
interacts
with
design
so
you're
going
to
least
bring
it
up.
A
Cameron
is
back
from
Microsoft
and
working
quite
a
bit,
and
so,
if
folks
have
design
work
related
to
Jupiter
lab,
please
loop
him
in
he's
available.
He's
definitely
can
help
out
and
I'm
also
having
him
start
to
manage
work
with
our
other
designers
sarika
and
sporty
as
well.
So
we
have,
you
know
a
good
amount
of
design
cycles.
It
folks
me
that.
A
And
looks
like
Sam
you're
next,
so
my
interest
sexually
dovetail
quite
closely
with
what
you've
been
speaking
about
over
the
last
few
minutes,
Brian,
which
is
that
we're
in
a
similar
situation,
where
I
support
a
whole
bunch
of
Statistics
PhDs,
who
are
trying,
interestingly
enough,
to
understand
the
use
of
a
complex
interface,
namely
Firefox,
and
so
we
have
large
amounts
of
data.
A
So
we
throw
it
on
s3
and
we
start
we,
you
know
a
hyper
competent
data
engineers,
pics
parkas
are
distributed
data
framework,
so
the
PhDs
got
to
use
spark
and
that
comes
along
with
Jupiter
notebooks
as
the
preferred
interface
and
immediately
they
were
all
covered
in
blood
from
paper
cuts
and
they
sort
of
slowly
climbed
the
learning
curve
and
I
I
really
speak
that
not
as
an
indictment
of
Jupiter
notebooks,
which,
for
this
wonderful
tool
that
have
grown
up
over
time.
If
anything,
the
spark
project
which
I
know
is
very
closely
tied.
A
A
Think
even
more
interestingly,
in
your
lead-in,
where
you
were
asking
your
colleagues
on
the
project
to
do
what
is
so
inartfully
called
dogfooding,
it
I'm
trying
to
figure
out
how
to
break
through
that
loop
and
the
method
I've
chosen
used
to
analyze
how
people
are
using
the
ID
to
get
information.
So
it's
not
about
when
I
look
at
it
and
I
use
it.
What
I
think
is
a
good
idea.
It's
imperfect!
A
Ironically,
that's
actually
the
work
that
we
are
all
doing
on
firefox.
So
the
reason
I
have
a
job
is
that
I'm
trying
to
that
for
Firefox
right.
So
this
is
all
very
long-winded.
I.
Thank
you
for
your
indulgence,
but
one
option
you
have
is
to
instrument
Jupiter
lab
in
beta
form,
and
you
would
have
to
you
know
there
needs
to
be
a
process
where
you,
let
users
know
hey.
A
We
would
like
to
collect
this
information
and
there
are
simple
things
you
can
do
like
you
can
say
they
created
this
kind
of
cell
this
kind
of
cell
this
kind
of
cell,
but
not
include
any
of
the
data
inside
the
cells
right.
Okay,
so
it
can
be
anonymised
in
that
way.
There
are
issues
with
fingerprinting
it.
You
can
make
that
as
complicated
as
you
want
to,
but
I
think
there
are
simple
practical
things
you
can
do.
A
You
see
where
I'm
going
with
this,
and
instead
you
can
just
say
they
opened.
They
opened
these
two
panels
and
immediately
they
got
a
trace
back
and
then
you
know
we
lost
connection
with
the
server
right
and
that
can
all
be
bundled
up
into
a
into
a
JSON
blob
and
sent
some
place.
That
can
be
later
analyzed
programmatically.
So
thank
you
for
listening
to
me.
Talk
about
all
this
Matthias
has
been
very
helpful
on
the
hacker
pad
about
helping
me.
A
A
Do
this,
but
here
are
20
people
working
on
desk
at
UC,
Berkeley
or
something
else
somewhere
else,
and
here's
all
their
work
in
progress,
stuff,
the
Messier
and
more
realistic
the
better
and
then
we're
going
to
release
a
small
tool
that
you
that
folks
can
use
internally
on
their
own
repositories
that
they
point
at
and
plug
in
arbitrary
functions
that
go
notebook
by
notebook
and
cell
by
cell
to
say
there
were
this.
Many
cells
with
trade
back
trace
packs
here
was
the
ratio
of
markdown
to
our
to
python
cells.
A
Here
are
all
the
notebooks
that
created
a
graph.
Here
are
all
the
notebooks
that
read.
Data
began
by
reading
data
from
s3,
then
put
it
into
a
distributed
processing
framework
and
then
did
X
with
it,
and
my
goal
is
to
generate
actually
a
notebook
with
some
of
this
information
so
that
we
can
start
to
speak
based
on
data
and
also,
hopefully
empowering
some
other
folks
who
aren't
necessarily
notebook
hackers
themselves,
but
they
can
take
an
example.
A
Notebook
pointed
to
their
own
repo
and
say:
hey,
you
know
what
you
think
that
the
important
thing
is
this,
but
I've
got
this
repository
of
600
notebooks.
That
says.
Actually
we
need
solve
this
other
problem
yeah.
So
if
anybody
has
suggestions
for
targets
and
again
it's
preferably
working
repos,
not
tutorials,
and
not
like
hey
here's.
My
magnificent
finished
project,
so
yeah
I'm
sure
Matthias
is
providing
really
good
idea.
A
Matthias
but
I
will
I
promise
to
go
back
and
read
them
and
drag
them
down
yeah
I
mean
what
one
thing
that
I'm
wondering
about
is
in
january.
I'm
teaching
a
data
science
course
for
undergraduates
in
at
cal
poly,
and
a
lot
of
their
homework
is
wouldn't
work
for
this.
In
the
sense
it
it's
very,
I
give
them
a
notebook
and
they
just
fill
in
the
cells
that
they're
sort
of
the
answers
to
the
problems,
but
towards
the
end
of
the
quarter.
A
A
It
may
I'd
have
to
think
about
sort
of
having
them
turn
in
notebooks,
or
you
know,
intermediate
notebooks
or
just
asking
them
to
run
a
tool
on
there
in
immediate
notebooks
or
like
that
may
and
in
that
or
I
think
that
user
group
would
be
nice
is
that
I
think
you'd
get
a
really
accurate
perspective
on
the
full
like?
How
do
you
start
a
large
data
science
project
from
picking
a
data
set,
exploring
the
data
set
cleaning
the
debt
like
it's
going
to
go
through
the
whole
thing,
and
you
also
have.
A
A
So
we
have
a
bit
more
background
on
instrument
instrumenting
and
user
testing,
so
we
have
done
some
qualitative
user
testing
on
the
I've
got
a
set
of
students
at
Cal,
Poly
or
designers,
and
also
CS
programmers,
and
we
had
them
do
user
tests
at
the
site
by
conference
this
past
summer
on
Jupiter
lab
and
we're
likely
going
to
do
more
of
that.
But
we
have
thought
about
doing
more
quantitative,
instrumentation
based
by
user
testing
in
Jupiter
lab
I.
A
Don't
think
it
would
be
difficult
to
write
a
plug-in
on
a
front
end
that
does
that
sort
of
instrumentation
the
easiest
way,
at
least
at
a
very
course
grade
level,
would
be
to
write
a
plug-in
that
records
events
that
happen
in
the
command
manager,
so
that
I
don't
know
how
familiar
you.
You
are
Sam
with
a
architecture
of
Jupiter
lab,
but
a
lot
of
the
top
level
actions
run
through
the
command
system,
and
so
you
know
we
simply
put
a
hook
in
there
that
every
time
the
command
is
called,
we
cord
that
event.
A
We
would
have
minimally
the
broad
workflow
of
everything
in
you
know
that
covers
sort
of
what
are
you,
what
files
are
user
users
opening?
What
are
they
doing
with
those
files
and
I?
Think
that
would
be
an
initial
point
and
there's
been
a
lot
of
discussion
about
how
to
deploy
instrument
instrumentation
in
an
open
source
project
in
in
particular,
Jupiter
and
there's
sort
of
two
answers.
One
is
deploying
it
for
all.
Our
users
does
bring
up
a
lot
of
reallys,
more
challenging
privacy
questions.
A
Obviously,
Mozilla
has
navigated
these
things
and
it
might
be
very
helpful
for
us
to
have
conversations
about
that.
The
easy
thing
that
we
can
do
is
that
a
number
of
us
have
fairly
decent-sized
deployments
of
Jupiter.
Where
we
are,
we
could
just
unilaterally
decide
on
that
deployment,
we're
going
to
instrument
it
and
like
at
my
university.
A
We
are
deploying
it
campus-wide
as
a
resource
of
research,
these
resource
or
the
faculty
and
staff,
and
then
also
within
individual
classes,
and
so
we
could
easily
run
instrumentation
over
the
period
of
a
quarter
or
six
months
and
gather
sort
of
detailed
instrumentation
data
in
general.
It's
much
easier.
There
are
many
fewer
thorny
issues
with
large
aggregates
on
than
there
are
with
individual
transmissions.
A
So
if
you
have
a
deal
where
you're
only
individually,
transmitting
within
illegal
and
sort
of
secure
entity,
and
then
what
it
publishes
is
aggregations,
that's
a
very
simple
way
to
avoid
a
lot
of
the
challenges
yeah
and
actually
so
we
have.
I've
talked
extensively
with
our
universities,
human
subject,
research
board,
and
it's
quite
interesting
as
long
as
the
data
that
we
lost
everyone
else.
D
A
Okay,
as
long
as
the
data
that
we
gather
is
only
used
to
improve
the
software,
it's
not
subject
to
a
human
subject
review
and
it
it's
it's
odd.
So
if
it
now,
if
we
were
to
gather
the
data
and
write
a
publish,
a
paper
about
it,
that
does
trigger
human
subject
with
you,
at
least
at
our
University
I
mean
other
companies
are
not
subject
to
the
same
review
processes.
Universities
are
but
like,
for
example,
for
me
to
instrument
our
campus
wide
Jupiter
hub
deployment
or
our
Jupiter
hub
deployment
for
a
data
science
class.
A
A
So
what
I
wondering
exam
can
you
speak
to
sort
of
what
fraction
of
the
paper
cuts
are
due
to
sort
of
Jupiter
in
general
versus
sort
of
the
inner
the
integration
between
Jupiter
and
spark?
The
latter
is
my
emphasis,
and
so
one
of
the
reasons
one
of
my
primary
motivations
for
engaging
the
community.
A
This
way
is
spark
is
not
the
only
distributed
computing
platform
out
there,
and
so
I
want
to
I
want
to
get
at
least
very
simple
information
on
what
is
idiosyncratic
21
distributed
computing
library
and
what
is
common
to
the
nature
of
distributed
computation.
A
So
I'll
give
you
a
simple
example:
has
a
cell
has
a
notebook
lost
contact
with
0
at
any
link
along
the
chain
between
the
notebook
to
the
HTTP
server
to
in
the
case
of
pi
spark
you
connect
with
a
Python
interpreter,
which
then
is
connecting
to
a
Java
VM,
which
then
is
distributing
out
a
thing.
So
there's
many
chains
along
the
way
and
things
can
break
at
each
point
and
beginning.
A
Waste
enormous
amount
of
time
waiting
for
a
job
that
is
never
going
to
return
because
of
that
arm.
Another
simple
one
is
some
operations
on
in
the
notebook
trigger
large
distributed
computations,
others,
don't
so
being
able
to
decorate
being
able
to
have
the
interpreter
have
a
a
common
language
for
saying
this
is
a
distributed
operation.
A
This
is
not,
then
there
are
workflow
issues,
so
a
lot
of
our
workflow
is
gather,
distributed
data,
bring
it
into
a
cluster,
do
stuff
with
it
and
then
either
pull
it
on
to
the
local
machine
or
push
it
someplace
else
and
I
did
not
know
about
Jupiter
lab
when
I
started
on
this
project,
which
I
thought
was
going
to
be
a
simple
small
thing
that
nobody
else
is
gonna,
be
interested
in,
but
I'm
excited
about
Jupiter
lab
because
there's
some
inherent
ways
in
which
the
architecture
that
is
going
to
make
us
writing
our
workflow
much
much
more
straightforward,
whereas
a
notebook,
it's
kind
of
like
how
do
we
rip
this
thing
apart
and
yeah,
so
there
there's
a
range
of
it
another
one
is
like.
A
Can
we
just
have
something
more
like
our
studio,
where
there's
one
panel?
That
shows
you
which
objects
are
currently
exist
in
your
workspace
and
in
turn
we
can
then
decorate
them
to
indicate
this
is
a
handle
to
a
distributed
object.
This
is
just
a
local
variable.
That
kind
of
thing
is
huge
if
you
are
unfamiliar
with
the
problem
space.
If
you
are
the
people
on
this
call,
that's
trivial
right.
A
But
I
should
note
that
particular
feature
like
a
variable
inspector
is
one
of
our
longest
and
most
common
feature,
requests
and
I
know
of
someone
who's
already
implemented
a
plugin
for
that
in
Jupiter
lab
awesome.
It's
not
it's
not
currently
open
source,
I,
I
plan
on
talking
to
them
and
saying
hey
like
a
lot
of
people,
would
really
really
like
that,
and
let's
see
if
we
can
get
it
actually
in
Jupiter
life,
but
it
even
if
they
don't
release
their
version
of
it.
A
I
think
that
is
something
that
we're
going
to
implement,
because
it's
awesome,
a
lot
of
users
would
find
it
very
very
helpful.
Don't
forget
you
had
mentioned
Mozilla's
experience
with
with
rules
around
collecting
user
data.
If
we
can
be
helpful,
please
just
reach
out
I'd,
be
happy
to
arrange
a
meeting
with
in
with
people
on
argh
on
our
data
policy
team,
fantastic
and
then
are
you
dared?
You
had
something
yeah.
C
I
just
wanted
to
ask
so
we're
not
talking
about
a
generic
instrumentation
library
that
we
could
plug
in
since
we
have.
This
notebooks
are
very
specific
format:
I'm,
not
familiar
with
those
tools.
Very
much
are
there
libraries
out
there
that
are
customizable
enough
to
do
this
myth
it
would
we
have
to
build
our
own
instrumentation.
A
The
short
answer
is
I,
don't
know,
but
I'm
sure
the
answer
is
yes
to
on
both
sides.
You
know
at
the
simplest
level,
you're
just
create
you're,
just
collecting
strings
like
and
the
string
could
be
here
are
the
nope
here.
Here's
here
are
a
list
of
cell
types.
A
You
know
markdown
cell
code
cell
code
cell
code
cell
code
sell
you
bundle
that
up
into
a
string
and
you
send
it
into
a
database
right
and
that's
just
a
you
know
a
row,
so
it
would
be
a
you
can
get
as
engineer
architect
as
you
want
to
so
mozillas
infrastructure.
Is
this
enormous
beast
with
hundreds
of
thousands
of
lines
of
code
but
mix
of
off-the-shelf
stuff
and
pure
custom
and
halfway
in
between?
So
it's
really
about
figuring
out?
A
What
is
the
right
tool?
You
know,
for
example,
in
Brian's
example,
where,
where
the
idea
is
that
you
have
a
local
installation,
that's
a
different
kind
of
approach
versus
a
tool
that
is
designed
for
the
Jupiter
project,
to
collect
information
from
beta
users
to
completely
different
needs,
and
you
would
evolve
the
project
in
different
ways.
That
answer
your
question.
Yeah.
C
To
some
extent,
I
guess
I'm
just
trying
to
figure
out.
I
guess
it
doesn't
matter
to
too
much
because
ultimately,
it'll
probably
just
be
some
plugin.
That
does
the
work,
irrespective
of
whether
that
plugin
is
implemented
using
and
off
the
shelf
library
or
if
we
have
to
write
it
ourselves.
But
I
just
I
don't
really
know
what
the
data
the
industry
is
with
those
off-the-shelf
tools,
because
I've,
never
all
the
tools
I've
ever
worked
on,
have
not
had
that
that
level
of
introspection
on
what
users
are
doing
in
an
automated
way.
A
A
Are
default?
Okay,
if
you
sorry
evaluated
this
one,
if
you
evaluated
the
spark
magic
yet
I
have
not.
Are
you
familiar
with
this
project
ur,
not
off
the
top
of
my
head
yeah,
so
I
kinda
yeah.
So
what
this
is
is
a
a
magic
function
or
Python
kernel
that
uses
the
rest
api
for
spark
livvie,
and
it
allows
you
to
hear
up
with
me
I'm
going
to
do
to.
D
A
One
thing
you're
talking
about
is
that
there's
a
very
clean
way
of
understanding
when
you're
running
code
in
the
Jupiter
Colonel
versus
one
year
running
code,
it
can
spark
because
it's
actually
a
magic
function,
so
our
syntax,
for
that
is
double
percent,
something,
but
don't
eight
percent
spark
or
dump,
and
it's
so
there's
at
least
some
sort
of
you
know
visible
artifact
for
the
user
of
saying.
Okay,
now
I'm
running
something
now
other
things
that
you're
talking
about,
like
being
able
to
monitor
their
remote
spark
jobs
understand.
Did
it
crash?
A
A
It's
just
that
it's
wedged
in,
and
so
the
question
I'm
interested
in
is
to
your
point,
different
parts
of
Jupiter
thinking
about
it.
What
are
the
broader
lessons
we
can
draw
as
a
user
experience,
and
certainly
something
like
spark
magic-
could
be
very
helpful
for
implementing
it.
But
I
done
I
want
a
dialogue
between
the
human
experience
and
and
kind
of
the
tool
shortcuts,
because
we
put
we
put
some
of
that
spark
integration
and
it
was
half
wonderful
and
half
useless
on
her
own
custom.
Notebooks.
A
A
A
So
are
great
and
that
this
is
directly
coming
out
of
other
people's
experience
in
supporting
spark
in
Jupiter
mm-hmm.
To
that
rate,
that
may
be
another.
I
will
mention
if
you
want
someone
who
wants
a
notebook
extension
that
directly
hooks
up
the
spark
ongoing
output
to
an
existing
notebook,
we
did
right
and
open
source
that
one
and
I
don't
think
we
publicized
it
much,
but
give
me
a
second.