►
From YouTube: Diagnostics WG Meeting - 2016-07-13
Description
A
All
right
welcome
everyone
to
the
nodejs
diagnostics
working
group
monthly
meeting.
It
is
July
thirteenth,
so
everybody
has
filled
in
their
names
on
the
attendee
list
on
our
agenda
today,
I
will
just
pull
out
and
if
every,
if
anyone
wants
to
add
or
subtract
I
think
we
should
start
with
a
sink
wrap.
If
there's
questions
about
that
or
things
to
discuss,
because
we
missed
it.
Last
time
we
have
Matthew
Loring
and
Ally
from
google
they're
going
to
talk
about
the
trace
controller
they've
been
working
on
I
put
down
inspector
protocol.
A
B
The
mainly
here
to
answer
your
questions:
if
anybody
has
anything
the
pp
is
there,
what
is
there
has
been
accepted,
I'm
still
working
on
the
public
API
their
little
ruts
that
make
it
painful,
but
it's
coming
along
and
we'll
begin
to
implement
those
in
pieces.
There's
one
major
change:
it
needs
to
be
made
to
be
internals
of
it
so
that
it
can
support
multiple
like
so
somebody
can
have
your
own
unique
set
of
hooks
instead
of
all
hooking
into
the
same,
the
giant
thing
and.
B
B
D
B
B
I'm
trying
to
see
if
there's
like
a
loose
wave
in
brainstorming
on
the
loose
weight.
I
like
still
allow
that
because
make
all
that
got
some
important
things.
If
you
don't
care
about
like
if
you're
just
calling
back
but
like
you,
have
a
domain
right
so
like
calling
back
in
c++
with
your
own
domain,
for
example,.
B
B
As
such,
it
wouldn't
it
automatically
not
run
like
the
next
step
q,
because
it
assume
that
you're
already
in
a
in
execution
safely
and
that
BAC
that
this
next
tick
you
would
be
run
for
you
automatically
after
it's
done
running
so
yeah,
some
minor
things
like
that.
So
basically,
you
will
just
cut
out
a
bunch
of
code
from
it
we'll
leave
it.
There.
D
B
Well,
it
should
be
I
mean
that
there'll
be
a
teeny
bit
of
duplicated
code.
Having
still
having
a
node
may
call
back,
but
like
the
amount
of
code
will
be
much
smaller,
like
all
the
all
the
hook
calls
into
pre
and
post
or,
and
then
those
things
also
the
to
clear
the
code.
Colombian
execute
all
that
will
you
remove.
So
it's
basically
just
left
on
my
head.
B
It's
a
setup
giving
if
there
is
one
call
the
callback,
teardown
domain
and
then
I
think
there's
something
to
with
how
our
homeland
will
work
in
that
case,
because
it
will
work
slightly
differently,
since
it
will
assume
it
can
be
caught
like
it
will
assume
that
there's
already
JavaScript
stack
on
top
of
it.
Oh.
B
D
It
just
so
familiar
with
it,
but
it
has
this
an
at
a
callback
ever
ever
whisper,
big
arms
collection,
so
that
will
sort
of
be
our
in
its
hook
in
Asian
grab.
And
then
this
rubber
object
has
a
core
method:
that
sort
of
PR,
pre
and
post.
Who
goes
you
and
then
eventually
you
will
destroy
this
rapper
who's.
Then
you
of
course
I'll
destroy
hook.
B
B
E
Was
it
or
work?
I
just
mean
more
currently
working
on
that
in
the
guy
working
loop.
We
actually
have
something
that
runs
like
Aldon,
an
example
we're
preventing
some
larger
modules
now.
So
it's
like
that
sounds
like
something
that
we've
got
a
factor
into
how
you
would
support
it
through
that
II,
you
know,
but
that
epi
as
well
yeah.
B
I'm
to
make
like
oh
I'll,
bring
some
that
too,
but
to
be
completely
honest,
I'm
not
sure
that's
feasible
has
to
as
some
of
the
things
that
does
is
completely
based
on
the
life
cycle
of
the
class,
which
is
why
you
hear
from
it
and
almost
you
to
like
magic.
We
inherit
firms
and
an
API,
then
I,
don't
see
how
that's
about
be
positive.
B
A
A
D
B
A
A
G
Is
this
better?
Yes,
right,
sorry
about
that
yeah!
So
so
there
have
been
some
exciting
developments
with
tracing
I
want
to
say
upfront.
A
lot
of
the
engineering
work
for
this
has
been
done
by
Raymond
who's
here
with
us,
as
well
also
on
our
team.
So
we,
the
the
tracing
controller
work
that
was
outlined
in
the
issue
that
I
think
I
saw
linked
in
the
minutes
is
I
has
some
components
that
will
live
in
v8
and
some
components
that
will
live
in
node.
G
So
the
there
is
now
a
pull
request
open
that
I
can
link
to
in
the
minutes
that
have
that
has
the
v8
component
of
this,
so
that
consists
of
a
mechanism
for
aggregating
traces,
filtering
traces
based
on
categories
and
also
all
of
the
interfaces
that
will
allow
node
to
specialize
how
tricks
are
collected,
add
custom
serialization
formats,
so
they
can
be
outputted
in
whatever
mechanism
is
most
desired.
I
can
link
down
here.
So
I
put
the
link
to
the
pull
request
for
that
in
the
minutes.
G
C
Think
that
would
be
up
to
like
having
that
that
could
be
a
follow-on
and
I
think
it
will
be
something
that
the
community
would
decide
on
how
to
best
expose
it
to
the
node
ecosystem.
I
think
the
goal
we
have
right
now
is
to
make
sure
we
have
enough
features.
We
have
it
increased
functionality
in
v8
available
and
that
is
so
plausible,
buy,
em,
better
and
and
and
that
that
we
have
a
basic,
a
sort
of
buffer,
trace
controller
and
buffering
mechanism
available
in
node.
So
that
note
core
can
get
traces
now.
C
A
G
So,
are
you
able
to
see
my
screen
right
now,
so
this
is
just
a
small
HTTP
server.
It's
going
to
read
a
file
whenever
it
gets
hit,
so
we
we
ran
this
and
put
some
load
on
it
and
it
emits
a
profile,
is
consumable
by
the
chrome
about
tracing
tab
and
you'll
get
a
view
that
looks
something
kind
of
like
this,
so
you
can
see
when
so.
This
is
something
that
we
wrote.
The
serialization
for
this
by
extending
the
interface
is
provided
by
v8
there.
G
It
could
be
extended
similarly
for
other
formats,
but
what
you
have
here
so
you
can?
You
can
zoom
in
here
we
see
at
startup
right
when
the
revenue
men
charged
starts.
We
have
v8
compiling
all
the
scripts
we
can
zoom
in
on
one
of
these.
We
have
some
compile
events,
it's
parsing
a
script
and
and
compiling
it,
and
then
we
see
there's
some
downtime
between
sit
here.
G
So
there's
some
downtime
between
the
initial
compilation
and
when
we
start
putting
load
on
the
server
and
we
see
the
note
event,
loop
is
sitting
idle
there
with
no
cpu
usage
and
then
we
can
come
in
here
to
kind
of
the
bulk
of
it.
You
see,
let's
come
back
here
for
a
bit,
so
what?
First
you
see?
A
few
of
these
red
tape
marks
dipping
down
below
the
purple
line.
G
These
correspond
to
GC
scavenge
events,
so
you
can
see
you
like
kind
of
the
frequency
and
the
time
taken
by
by
GC
and
additionally,
if
we
kind
of
zoom
it
around
this
area.
So
we
see
that
interleaved
among
the
the
main,
the
main
event
loop
execution.
There
are
a
few
of
these
file
system
operations.
You
can
get
information
on
how
long
specific
ones
took
by
clicking
in
and
also
you
can
highlight
over
a
region
of
this
graph
to
get.
C
A
C
Well,
it
has
a
time
stamp.
I
think
it
also
has
CPU
time
in
it
and
I
think,
depending
on
the
types
of
type
of
trace
point,
you
have
there's
lots
of
types
there's
once
for
synchronous
events,
others
for
asynchronous
events,
and
you
can
add
additional
metadata
along
with
the
trace
points.
Do
with
you
it's
up
to
you,
how
you
want
to
use
the
trace
points.
G
G
So
serialization
there's
a
trace
writer
interface.
You
can
special
node
will
be
able
to
write
specializations
of
the
trace
writer
interface
to
output,
in
whatever
format
to
whatever
file
yeah.
That's
all
that
there's
a
general
interface
and
we've
implemented
one
specific
instance
of
this
to
output,
the
chrome,
consumable
format
to
a
file
called
node
node
trace
log,
or
something
like
that.
But
that's
that's
completely.
Yeah!
That's
configurable!.
A
G
E
A
C
G
C
G
The
one
thing
that
we
need
to
keep
so
on
the
timeline
we
need.
These
changes
are
just
landing
in
v8
now
and
there's
going
to
be
lat
lead
time
before
v8.
So
this
will
probably
land
in
v8
5.4,
hopefully,
and
so
there's
going
to
be
lead
time
before
that
gets
picked
up
by
node.
So
we
will
continue
working
out
the
node
prototype,
but
it
will
have
to
be
on
a
branch
of
on
a
branch
of
node
that
has
v8
5.4
in
order
to
integrate
with
v8.
A
So,
for
example,
can
the
agent,
instead
of
loading
up
v8
inspector,
can
it
load
up
some
sort
of
Inspector
interface
so
that
there
could
be
alternate
back
ends
in
the
future
and
I
also
proposed,
and
this
one
is
probably
less
important,
but
separating
that
that
protocol
gateway,
which
is
generic
from
the
dispatchers
to
the
v8,
which
is
of
course
mapping
from
the
incoming
protocol
to
what
v8
expects?
So
that's
really
a
secondary
thing
like
we
could
have
that
protocol
gateway
be
a
little
more
generic
and
shared.
C
A
Okay,
so
the
other
thing
I
was
going
to
ask
about
on
inspector
is
I
know
Ali.
We
one
of
the
things
is
that
we
need
a
CLI
debugger
for
inspector
for
that
protocol.
I
think
there's
been
a
couple
other
things,
I've
noticed
in
dreads
of
tasks
that
node
can
do
or
enhancements
that
we
might
want
to
pursue
after
things
are
stable
d?
Is
there
could
we
start
building
a
list
of
those
kinds
of
things?
Do
you
have
other
things
elite.
C
A
F
F
As
well
at
least
three
feet,
four
five,
okay
yeah,
so
I
saw
somebody
else
right
on
the
on
the
issue.
It's
issue
number.
What
does
it
say
here,
58
I
heard
somebody
say
as
well
that
they
could
not
make
it.
Also,
how
many
people
are
you,
so
the
question
is:
should
should
we
postpone
it
or
I
mean
of
Orchard
Beach?
We
have
to
be
headed
there.
I.
C
F
So
I
taking
a
look,
look
at
what's
what
no
related
conference
is
that
our
international
are
coming
up.
Besides
no
summit
and
I
can
see
the
to
note
is
interactive
the
new
year,
one
and
North
American
one,
and
then
there
is
a
note
con
tu
as
well.
That's
kind
of
like
the
only
ones
that
I
think
are
big
enough
to
attract
all
the
key
people
are.
F
F
So
basically,
it
all
started
at
no
content
venture
where
we
had
a
session
about
tracing
an
ATM
occasion,
performance
management
arm-
and
we
kind
of
came
to
the
conclusion
that
it's
really
hard
to
do
a
note,
and
there
was
a
lot
of
the
key
mpm
sorry
APM
APM
then
does
at
know
couple
of
inches.
So
we
had
a
really
good
talk
about
how
how
we
all
solving
basically
the
same
thing
over
and
over
again
so
first
off
it
might
be
a
good
idea
to
basically
share
some
things.
Second
off.
F
Maybe
some
of
this
could
make
its
way
into
node.
Maybe
not,
but
that's
that's
the
discussion
that
needs
to
be
he
had
so
so
that
was
basically
idea
of
the
PM
summer.
They
kind
of
akin
to
the
there
was
an
error
summit
as
well.
Oh
this
year.
I
think
it
was
so
it
was
kind
of
in
that
spirit
that
you
wanted
me
to
make
it
so
yeah.
F
That's
that's
a
very
brief,
very
brief
overview
and
the
people
who
were
there
was
basically
a
Panetta
myself
from
upbeat
there
were
who
were
the
others
out.
There
was
I,
think
note
shows
with
as
well,
and
there
was
I
forgot
his
name.
There
was
a
guy
from
New
Relic
as
well.
He
was
not
working
on
the
under
on
the
agent,
but
you
wanted
to
to
ping
the
right
people
at
me
really
as
well.
He
was
he
was.
He
was
participating
in
the
discussion.
Also.
A
F
F
So
you
saw
or
other
things
well
yeah,
so
so
I
think
its
traces
on
another
level.
Then
we've
been
talking
about
so
far
at
this
meeting,
it's
more
application
level,
traces
usually
related
to,
but
not
only
HTTP,
requests
and
and
that
the
database
request
and
IO
requests
that
goes
on
during
an
HTTP,
you
cuz,
so
basically
what
you
get
from
from
New
Relic,
for
example,
but
but
also
it's
sometimes
you
want
to
instrument
your
own
code.
F
You
have
some
kind
of
long-running
function
that
you
want
instruments,
so
you
might
put
a
start
trace
and
trace
around
that
function
of
stuff
like
that
Manuel
traces.
So
to
speak,
so
it's
a
more
high-level
tracing
than
the
stuff
that
we
usually
talk
about.
But
it's
it's
it's
it's
a
big
problem
for
breaking
windows
to
get
the
to
get
the
right
hooks
into
note,
and
today
what
you
need
to
do
to
figure
out
how
long
does
an
HTTP
a
state
which
database
queries
are
occurred
during
doing
this,
HTTP
requests,
etc.
F
You
need
to
do
a
lot
of
really
ugly
monkey
patching
and-
and
we
of
course,
are
looking
into
all
the
nice
stuff
that
Asian
grab
is
adding
to
this
are
giving
us
here
which
will
help
some
of
it,
but
not
all
of
it.
Also,
there
is
a
big
issue.
Right
now
will
relate
it
to
the
years
modules,
where
the
way
that
we
make
this
work
is
that
we
basically
monkey
patch
the
underscore
look
function
on
the
module.
F
Module
are
to
basically
cook
into
all
the
modules
that
have
been
loaded
and
inject
our
own
code.
I,
don't
racing
logic
into
those
modules
sounds
really
awful
any
kind
of
this,
but
of
course
we
can't
do
that
with
year
six
modules,
because
it's
it's
not
possible
to
modify
the
imported
modules
near
60.
There
is
also
an
issue
that
we
need
to
discuss.
I
know
a
lot
of
like
news
find
people
are
discussing
this.
It's
not
gonna
notice.
Luckily,.
A
I
was
helpful
for
me.
Matt
and
Ali
does
the
the
tracing
controller
with
that
if
we
were
able
to
do
a
user,
the
user
side
implementation,
so
that
you
know
we
could
be
used
from
within
modules?
Would
that
be
a
good
place
to
start
for
this,
or
do
you
think?
That's
it
really,
that's
the
wrong
place
to
start
there's.
C
Slightly
different
concerns,
so
so
the
problem
the
promise
was
talking
about
is
more
related
to
application
level
understanding
so
so,
and
it
has
to
do
a
little
bit
with
the
tracking
of
context
across
asynchronous
flows,
so
so
so
sure
the
trace
points
could
be
ultimately
the
tracing
that,
with
the
trees
points
that
we
are
working
on
with
trace
event
could
be
ultimately
the
source
of
information
that
a
PM
renders
could
use.
But
that's
not
really
the
problem.
C
C
C
So
right
now
it's
okay!
The
answer
is
yes,
that's
the
idea.
It
should
be
cheap
enough
that
you
could
enable
it
in
in
production.
I
mean
it
depends
on
how
many
events
I
mean
it's
right.
Now,
it's
designed
for
performance
around,
so
it's
not
designed
for
so
it's
designed
for
when
you
are
actually
going
to
go
investigate
conversation,
but
it
depends
on
the
number
of
events
you're
going
actually
going
to
be
tracing.
C
So
if
you're
tracing
every
viet
execute
that
happens
millions
of
times
a
second
and
if
you're,
going
to
trace
that
that's
going
to
have
a
cost,
so
it
depends
on
the
type
of
thing
you're
interested
in
tracing.
So
I
look
at
this
like
Tracy
minstrel
tracing
as
a
slightly
different
kind
of
a
beast
which
has
different
type
of
information.
Think
of
it
as
structural
profiling,
whereas
the
the
EPM
style
tracing
is
more
about
I'm,
going
to
take
a
sample
of
some
of
your
HTTP
requests
and
for
these
HTTP
requests.
C
I'm
going
to
tell
you
all
the
our
pcs
that
happened
as
during
this
HTTP
request.
So
so
it's
possible
that
you
could
use
trace
event
data
for
for
the
latter
as
well,
but
I
mean
they're
coming
at
the
problem
from
two
different
perspectives.
One
is
like
very
high
level.
Other
one
is
as
low
level
as
it
gets
and.
A
C
C
The
APM
tracers
are
not
interested
in
those
fish
points
today,
because
that's
too
much
detail
but
down
the
road
you
would
imagine
the
node
would
have
some
trace
points,
so
we
could
say:
oh
yeah,
we
I
I,
started
in
a
festive
read
here
and
I
find
it
here,
and
that
may
be
something
that
IBM
vendor
could
be
interested
in,
or
there
may
be
other
things
that
the
AP
manners
are
interested
in
and
maybe
further
down
the
road.
Maybe
more
people
would
add
in
the
ecosystem.
C
F
What
we
have
talked
about
is
if
it
makes
sense
to
make
some
kind
of
generic
protocol
that
let's
say
your
database
vendor,
and
you
have
a
note
driver
that
that
you
that
you
maintain
it
would
make
sense
for
you
to
build
into
that
this
generic
APM
tracing
protocol.
So
you
know
that
you
work
out
of
the
box
with
all
the
major
APM
vendors.
So
maybe
maybe
there
is
so
we
always
talk
kind
of
down
that
road.
So
maybe
this
would
kind
of
be
the
protocol.
F
D
F
Lose
it
well,
it
solves
the
the
issue
about
all
the
I
mean
if,
let's
just
imagine
that
we
get,
the
promise
is
a
process
that
next
take
and
the
promises
and
all
that
stuff
as
well,
then
it
solves
all
the
core
stuff.
Are
the
girls
inside
that
goes
on
inside
note,
but
you,
for
example,
the
postgres,
the
p
ET
module,
for
example,
have
an
internal
call
back
you
that
does
not,
because
it
maintains
a
connection
pool,
of
course
to
the
database.
So
will
you
make
a
query?
F
Your
callback
is
actually
cute
in
an
array,
and
it
just
shoots
your
query
down
down
the
socket
two
on
one
of
the
one
of
the
open
sockets
and
when
it
gets
back
a
result
from
database
it.
It
looks
in
the
callback
unco
which
wich
callback
should.
I
call
a
based
based
on
this
result
that
I
got
back
so
and
that
callback
use
a
portion
utilized
right
when
you
make
the
first
queries.
So
if
it's
still
in
that
context,
so
so
you
can
no,
which
request
was
actually
responsible
for
calling
that
call
back.
C
Yeah,
so
so,
ultimately,
the
problem
is,
you
think,
wrap
the
purpose
of
facing
crap
is
to
give
you
insights
into
a
sink
flow
within
note
chord,
and
the
problem
is
so
so
we
are
trying
to
track
context
and
context
can
be
broken
by
any
user
space
library.
That
takes
a
call
that
puts
it
in
a
queue
and
then
calls
it
later
in
under
a
different
context,
and
that's
basically
the
problem
that
I
think
rap
is
not
designed
to
solve.
I
mean
domain
sort
of
solved
it
in
the
past.
C
It
they
weren't
fully
designed
and
say
that
that's
not
giving
the
problem
that
the
notion
of
logical
current
context,
the
concept
is
there
in
dotnet,
for
example,
I
think
each
C
sharp
has
this
concept
of
a
logical
call
context.
That's
what
we're
trying
to
propagate
and
us-based
libraries
can
break
it
very
easily,
and
a
and
E
still
need
to
monkey
patch
userspace
libraries
to
propagate
context
on
connection
tools
and
around
other
cues.
F
D
That
is
in
you,
but
I
think
in
the
long
term.
We
eventually
want
a
shoe
rack
to
solve
this
issue
such
as
native
add-ons
does
not
break
and
yeah
yeah
the
context.
However,
you
how
we're
gonna
do
that
is
not
really
clear,
I
mean,
hopefully
the
AC
web
class
that
Trevor
talked
about
would
be
enough,
but
I
mean
there
will
always
be
modules
that
does
not
behave
nicely
and
still
yeah.
Those
will
be
an
issue,
but
hopefully
we
can
stall
them
note.
C
C
F
He
yeah,
I
have
I've
made
a
duel
that
I
will
I
will
send
out
or
like
she
just
added
posted
a
comment
on
the
on
the
issue.
I,
don't
know
if
there
is
I,
don't
want
to
cut
off
the
discussion.
Anything
but
I,
don't
know
if
there's
any
more
questions.
Anything
like
that
that
we
need
to
discuss
here
on
the
call.
F
F
F
C
Think
might
be
what
sort
of
proposing
one
of
the
other
sort
of
European
conferences
and
seeing
if
people
are
able
to
make
it
and
if
not,
then
then
he
has,
we
can
postpone
it.
I
think
it
I
mean
this
is
an
important
issue
for
the
ecosystem
and
for
the
overall
completeness
off
I
mean
diagnosed
heading
for
nodejs,
so
it
will
be
good
if
people,
if
different
AP
lenders
could
come
to
the
same
way
on
how
to
trace
no
absolutely.
F
F
F
F
Yeah
there
is
that
step
222.
Following
days
after
after
the
conference,
I've
been
thinking
a
lot
about
that
and
I.
Don't
know,
that's
a
good
thing
or
bad
thing
I
like
if
people
are
going
to
be
there
because
of
that
or
if
the
people
want
to
go
to
both
and
then
of
course
have
to
choose
and
we're
gonna
be
a
conflict.
F
We
will
see
what
the
word,
what
the
duel
gives
us
I'll
make
a
note
of
it.
So
people
are
people,
know
that
it's
it's
it's
a
choice
or
one
of
the
other.
If
they
we
put
it
at
at
the
nokia
syntactic
europe
at
least
we
put
it
after
I've
heard
that
it
it's
a
bad
idea
to
put
it
the
day
before
no
Jess
interactive
Europe.
F
If
because,
apparently
is
not
possible
to
to
get
into
the
venue
the
day
before,
of
course,
you
could
do
it,
some
someplace
totally
different,
but
there's
something
about
being
a
holiday
in
Amsterdam
or
something
the
day
before.
That
means
that
hotels,
I,
extremely
expensive
and
stuff
like
that.
So
so
that's
why
the
day
before
probably
is
not
so
good.