►
From YouTube: Diagnostics WG meeting - May 6 2020
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).
B
B
C
C
F
C
E
Exactly
and
it's
significant
so
that
it
you
know,
people
don't
do
their
stuff
and
then
just
turn
it
on.
It's
like.
Oh,
wait,
a
second.
What
happened
because
this
is
improving
the
perform
like
is,
is
it
today
as
long
as
you
don't
turn
that
on
well
I
guess,
even
today
you
have
the
the
performance
hit,
no
matter
what
right
yeah.
C
Well,
there's
like
promise
some
class
process
wraps
up
class
of
that
which
it'll
wrap
around
every
single
promise,
I
think
always
and
every
single
time
it
does
that
it
does
JC
tracking
for
that
object
as
well
right
now,
there's
a
bunch
of
extra
overhead
to
all
that
GC
tracking
and
stuff.
So
what
what
this
instead
does
is
it
just
puts
two
simple
properties
on
promise:
the
casing
ID
and
the
trigger
ID,
and
that's
basically
it
so
it
just
does
all
of
that
on
the
JavaScript
side
instead,
so
can
bypass
creating
the
icing
craft
entirely.
E
G
C
Yeah
yeah
haven't
really
considered,
but
the
impact
will
be
on
domains
the
the
way
it's
designed,
though
it's
will
only
use
this
fast
path
mode
if
there's
no
just
our
hook.
If
there
is
a
destroy
heck,
it
just
uses
the
existing
code,
so
it
won't
hurt
the
performance
any
more
than
it
already
would
have
been,
which
is
good.
E
G
So
a
ten
and
what
I
want
to
say
is
users
may
enable
the
slow
mode
of
awesome
cook
so
that
an
existing
one
not
directly.
They
may
do
it
indirectly.
It's
not
obvious.
If
you
document
it.
If
you
use
this
joy
hook,
then
it
will
be
faster.
If
you
don't
use
it,
it's
faster
because
there
may
be
some
other
user
and
the
system
which
verses
you
back
to
the
slow
mode.
E
C
C
H
C
C
B
B
I
B
I
Yeah
I
think
with
respect
to
the
discussion.
We
have
some
amount
of
consensus
around.
When
should
we
change
the
report
version,
which
is
basically
when
the
structure
of
the
data
format
changes,
but
then
the
contention
is
about.
What
do
you
mean
by
structure
of
the
JSON
data?
Is
it
just
that
the
schema
change
or
the
key
change
value
change?
There
are
all
sort
of
combinations.
If
you
look
at
the
data,
the
the
key
can
change,
the
value
can
change
and
the
structure
can
change.
I
B
B
Okay,
so
moving
to
the
next
one
proposal
to
drive
Diagnostics
working
group
initiative,
proud
user
John
journeys,
we've
been
working
on
that
here,
deep
types
and
since
last
week
we
are
alternating
between
meetings
and
the
types.
So
we
are
meeting
every
week
and
one
meeting
why
the
other
week
we
we
are
deep,
diving
on
issues.
B
K
K
I
I
I
forgot
what
are
those
the
original
discussion
and
all
that
it's
possible
that
some
of
the
items
which
are
discussed
and
kept
in
the
user
journey
document
is
not
converted
or
translated
into
the
documentation
in
the
repo
did
we
have
a
conversation
around
that
I
mean
I
just
wanted
to
list
and
what
we
do
the
deep
dive.
What
is
the
next
step.
I
A
So
I
think
it's
more
like
the
debugging
phases
so
like
basically
post
feed
how
to
do
the
tutorial.
Definitely
not
the
part
about
what
to
improve
I.
Think
that's
mostly
for
our
group,
but
I
think
there
is
a
value
in
sharing
with
the
broader
community.
Just
like
what
are
the
tools.
The
recommendations
for
the
use
case
is
how
to
use
them.
I
A
I
am
open
to
suggestion,
I
mean
the
problem
is
again.
This
is
a
working
group,
so
we
can't
really
assign
us
to
people
so
I
guess
our
main
goal
could
be
just
identifying
the
gaps
and
the
work
that
could
be
done
to
improve
and
maybe
kind
of
pointing
out
some
priority,
but
I
yeah
I
think
that's
this
output
that
we
can
make,
and
hopefully
people
will
be
interested
and
excited
to
work
on
them.
Yeah.
I
Yeah
make
sense,
yeah
I
just
just
wanted
to
create
with
that,
because
there
are
some
newcomers
in
the
working
group,
so
they
they
should
not
feel
under
the
impression
that
okay,
the
discussions
and
the
content
that
is
happening
in
the
deep
dive
is
implicitly
assigned
with
somebody
or
things
like
that.
That's
not
the
case.
It's
open
for
anybody
to
pick
it
up
and
convert
that
into
the
best
practice
documentation
at
the
conveniens
matching
skill
set
interest
and
bandwidth
yeah.
A
But
I
think
you're
making
a
good
point
that,
like
maybe
it
would
be
time
to
just
document
the
whole
like
how
we
are
doing
a
deep
dive
like
what's
the
process
for
that.
What's
the
goal,
what
we
hope
from
it,
because
it's
like
kind
of
living
in
our
mindset
but
I,
don't
think
it's
a
commented
anywhere.
Yeah.
I
A
A
D
K
B
C
Yeah
have
been
working
on
that
Venables
has
issue
with
a
single
weights
and
from
you
away
the
babble
there's
kind
of
this
weird
dead
zone
that
Mason
cooks
will
have
like
zeroed
out
IDs,
no
I
think
reelers
back
within
the
venable
like
it.
If
you
using
like
Mason
political
storage
like
before
after
the
weights,
you
don't
have
the
storage,
but
inside
of
it
you
won't
know
so.
I'm
kind
of
that
breaks
literally
every
APM,
but
I
found
a
way
to
fix
it.
C
C
But
basically
the
problem
is
that's
Venables,
but
with
promise
hook
it
is
attached
directly
to
the
actual
promise
class.
So
if
your
injury
using
something
that's
promise
like
but
not
actually
promise,
it
never
omits
promise
of
events,
but
it
it
does
get
upgraded
to
a
promise.
But
but
when
you
do
the
awaits
or
if
you
like,
handed
venable
into
promise
dug
resolved,
it
will
go
through
like
an
internal
promise,
upgrade
step
to
turn
it
into
dual
promise.
C
But
there's
kind
of
this
like
two-phase
thing:
if
there's
like
powder
there's
like
the
outer
promise
like
the
weight
itself
and
then
the
inner
promise,
that
upgrades
is
actually
inside
a
micro
micro
task,
which
is
unwrapped,
but
it
has
no
basic
hoax
context
at
all.
And
so
that's
the
problem
is
there's
currently
nothing
linking
these
two
promises
together.
So
the
conduct
just
disappears
so
I'm
trying
to.
H
C
Out
a
way
to
find
close
together,
so
we
can
sort
of
detect
it
and
I'm
kind
of
working
on
how
to
come
in
an
actual
code
right
now.
But
basically,
if,
if
you
have
the
net
events
and
it's
followed
by
promise,
resolve
within
the
same
execution,
I
think
I
D,
then
that
means
that
it
is
immediately
resolved
promised
so
either
like
how
much
that
resolved
or
amenable.
C
As
I
said,
I
want
to
make
a
lower
level
version
the
the
reason
why
I
want
to
get
some
items.
None
of
those
there's
been
a
lot
of
pushback
thing,
not
just
for
the
potential
overhead
of
it,
but
also
an
argument
that
it's
not
needed
because,
like
async
resources
exist,
which
I
I
can
sort
of
partly
understand.
Why
why
people
would
think
that
argument?
C
Just
adds
this
extra
cognitive
burden
to
anyone
that
wants
to
write
a
node
module
for
anything,
because
you
either
have
to
be
aware
of
just
a
potential
problem
in
any
Asian
code
or
risk
breaking
like
the
entire
application
like
it.
It's
it's
not
so
I
guess
there's
a
lot
of
things
to
be
aware
of
in
node
that
early.
This
is
the
best
practice
or
something
like
this.
But
this
is
a
sort
of
thing
where,
if
you
are
not
aware
of
this,
you
you
can
literally
bring
some
of
that
mm-hmm.
C
It
is
it's
not
obvious
that
it
should
be
relevant
to
your
code,
so
as
much
as
possible.
We
should
try
to
avoid
relying
on
a
sand
cooks
to
fix
these
problems
and
should
try
and
fix
them
ourselves.
If
we
can
just
trying
to
trying
to
figure
out
how
to
explain
this
in
the
issue
and
get
get
a
little
more
buy-in
on
like
guess,
we
could
actually
do
something
about
this.
C
Possibly
I
assume
it
would
be
faster,
but
it's
it's
hard
to
say.
I
I,
don't
have
a
completely
clear
idea
of.
What's
the
be
a
fix,
would
look
like
but
I,
like
kind
of
some
idea,
what
it
might
look
like.
G
C
The
problem
of
fixing
it
on
the
v8
side,
those
I,
got
I've,
taken
this
issue
to
the
VA
team
before
and
they
basically
said
we
don't
have
the
time
for
that.
We
don't
care
enough
about
it.
We're
not
gonna
do
anything
about
it.
If
you
want
to
fix
it,
fixing
yourself
so
like
we
could
put
a
bunch
of
time
into
trying
to
fix
it
ourselves.
But
that
means
it's
v8
code
that
we
have
to
maintain.
C
C
B
C
C
B
G
Maybe
two
things
here:
on
the
one
hand,
I
really
just
didn't
get
a
dish,
not
fixed
I,
agree
that
we
should
do
it
not
to
move
it
to
userland,
and
there
was
regarding
the
stable
API
in
housing
cooks.
There
was
some
initiative
from
vladimir
to
move
Osment
local
storage
class
of
this
experimental,
so
not
the
complete
ax
as
in
hooks,
because
they've
expose
Indian
arts
even
for
we
count
more
less
expose
it.
A
thousand
local
storage
class
could
be
exposed
and
and
move
sort
of
experimental.
So
there
is
some
movement
in
this
area.
C
E
G
C
Yeah,
the
original
thought
was
to
give
it
its
own
namespace,
but
what
we
could
do
is
just
make
it
like
move
it
out
of
experimental
in
the
current
namespace
and
then
like.
We
could
create
another
top-level
module
just
like
react
sport
and
then
eventually
down
the
line
somewhere
just
click
and
move
because
from
one
place
to
that
other.
If
it's
like
reacts,
portage
that
we
could
just
like
to
move
the
react.
Sport
just
like
swap
places
right.
E
C
C
D
E
E
J
H
Sorry,
x64!
That's
really
what
I
mean
before
I
know:
mateus,
you
put
a
PR
into
unofficial
and
it's
not
clear
to
me
so
because
I
don't
know
the
internals
of
sort
of
know
how
the
structures,
how
builds
get
promoted,
sort
of
have
the
possibility
of
getting
that
into
a
sort
of
an
official
release
or
some
sort
of
release.
H
H
Okay,
we
people
nod
their
heads
when
they
can
see
things:
yeah,
coop,
fantastic,
okay,
so
very,
very
brief
overview
of
what
is
he
be
PF
n,
so
EBP
F
is
stands
for
extended,
Berkeley
packet
filter
and
you
will.
It's
also
been
shortened
back
now
to
be
PF.
Okay,
so
then,
now
you
sort
of
interchangeably
but
based
on
the
technology
that
was
introduced
in
nineteen
nineteen
90s
into
freebsd
into
the
rest
of
the
eunice
YZ.
H
It's
definitely
sort
of
comparable
features
in
terms
of
what
it
does
for
the
colonel
so
allows
you
to
write
little
complete
programs
that
you
can
embed
directly
into
say,
if
safely
in
bed
directly
into
the
colonel
okay,
so
very
quickly
sort
of
the
runtime
architecture,
then
so
you
have
a
registers
set
of
registers
and
a
stack,
and
then
you
have
a
pool
of
data
storage
underneath
that
they
referred
to
as
Maps.
Okay.
H
So,
like
I,
said
it's
a
very
simple
sort
of
reduced
instruction
set
computer
and
then
in
order
to
get
a
program
into
this,
you
start
off.
A
user
would
write
a
program.
This
could
be
a
Python
script.
This
could
be
a
Russ
program.
This
could
be
a
C
program
and
this
can
be
actual
byte
code.
But
what
happens
is
that
the
program
generates
a
set
of
ebf
back
cold,
that's
verified
by
a
static
verifier
before
it's
loaded
into
the
kernel.
H
The
VBA
runtime
then
runs
the
actual
application
and
attaches
your
little
BPF
program
to
whatever
prose
that
you
specified
when
you
define
the
program,
so
the
bits
that
I
I'm
interested
in
terms
of
notice,
we
have
USD
T's
specified
for
TCP
for
HTTP
for
GC,
and
we
can
attach
programs
on
to
those
events
within
no
Jess
and
Mateus
did
a
brilliant
breakdown
in
the
blog
posts.
Sorry
in
the
in
the
issue
of
all
the
other,
what
K
probes
do
what
you
probes,
do
and
trace
points?
H
Okay
and
then
sort
of
outside,
so
you've
got
this
little
program
running
in
the
kernel
and
that's
attached
to
a
set
of
probes.
It's
generating
data
into
this
f
MA
and
then,
as
it
then
fires
events
out
of
that
out
of
the
kernel
into
the
your
user
lamps.
So
you
can
see
those
events
or
you
can
let
those
events
group
into
a
set
of
statistics
and
then
take
them
out
as
an
ASIC,
read,
okay
and
then
just
in
terms
of
the
script.
Fine,
just
closing
off
on
this.
H
So
in
terms
of
the
script
ecosystem,
you
I
mentioned
that
you
have
the
ability
to
write.
Python
Python
Ross
go
et
cetera,
et
cetera
in
order
to
generate
those
scripts
there's
also
a
sort
of
a
dtrace
type
language
called
bps
race
is
very,
very
similar
to
dtrace
in
terms
of
the
how
it
how
it
works
and
it
syntactically,
and
so
you
have
those
two
options:
there's
an
external
library
called
libby's
GCC
that
can
you
can
use
to
integrate
with
ebps
ecosystem,
and
then
you
can
write
the
byte
code
itself,
any
BPF,
okay.
H
So
that's
a
little
conscious
of
time,
but
that's
a
whirlwind
tour
of
what
sort
of
EBP
F
is
and
why
I'm
interested
in
introducing
it
into
the
or
in
trying
to
get
a
build
for
Linux.
So
we
can
have
this
runtime
real
time,
parole,
attachment
to
know
Jess,
that
we
can
build
diagnostics
and
other
interesting
things
off
the
back,
oh
not
just
in
the
node.js
layer,
but
how
the
no
jess
interacts
with
underlying
pieces
of
the
infrastructure
like
the
operating
system,
disk
network,
etc,
etc.
Alright,.
E
H
E
Like
if
your,
if
your
kernel
is
not
new
enough
to
actually
use
the
probe
successfully,
then
it's
like
okay,
that
that
can
just
be
a
limitation,
it's
more
like
if
you,
if
the
binaries
are
gonna,
now
not
work
or
be
a
problem
for
people
who
don't
have
the
library,
that's
the
different,
the
kernel
level,
that's
a
different
thing,
because
I
think
it's
gonna
be.
We've
talked
a
whole
bunch
of
different
times
about
like
additional
node
binaries
and
there's
a
strong
resistance
to
that.
E
Just
because
of
the
extra
work
in
terms
of
the
you
know
the
build
flow
of
building
additional
binaries
confusion
among
customers
or
which
ones
they
can
use
so
the
most
straightforward
path
to
get
something
that's
like
in
the
official
binaries
is
if
it
could
just
be
on
without
affecting
anything
like
if
it
slows
thing
down.
That's
another
concern,
but
if
it's
just
not
gonna
run,
that's
just
certainly
concerned,
but
if
it's
kind
of
like
this
can
be
turned
on
and
there's
no
negative
impact,
then
that
is
gonna
be
a
lot
easier.
Yeah,
okay,.
B
Yes,
the
probes
are
just
annotations
on
the
binary
that
are
not
executable
exactly
as
well
as
a
few
extra
call.
Paths
for
those
cobwebs
are
not
something
specific.
It's,
for
example,
tracking
the
garbage
collector
as
for
performance,
the
way
us
city
probes
are
you
I
think
they
slide
overhead,
even
when
they
are
disable
it
on
OGS.
B
B
B
E
B
B
E
Right
and
I
guess
just
the
last
thing
was
the
last
a
couple
years
ago,
probably
now
I
think
you
know
these
kind
of
traces
were
discussed,
but
there
wasn't
really
any
champions
for
them
and
and
that's
kind
of
why
nothing
has
happened
right.
There
hasn't
been
anybody,
who's
been
sort
of
committing
to
keep
them
working,
tested.
K
E
H
So
for
the
country,
a
nod,
full-time
developer
in
development
full-time,
so
I
can't
want
to
be
cautious
about
how
much
have
commits
we're,
definitely
interested
in
helping
helping
with
the
performance.
Metrics
gather
that
detail
and
we're
where
I
can
and
support
in
whatever
additional
effort
needs
to
go
in.
So
we
have
HAP's
in
terms
of
the
the
features
or
the
work
that
needs
need
to
be
done
previously.
For
the
freebsd
side,
it's
been
when
there
has
been
changes
in
some
of
the
some
of
the
operating
system
features.
H
So
the
the
way
that
the
way
that
arguments
were
being
managed
within
the
DJ's
implementation
improved
between
eleven
and
twelve.
So
we
we
have
to
zoom
back
in
there
and
sort
of
patch
those
in
and
then
an
Olfa
door.
Hatsue,
look
at
some
of
so
how
the
the
probes
are
generated
before
the
mangled
back
into
the
executable
Fodor
had
to
look
at
that
about
twelve
months
ago
as
well.
H
B
J
E
B
H
H
B
E
Right
now
my
question
would
be:
is
it
like
not
necessarily
the
end
JavaScript
developers,
but
the
people
responsible
for
production
implementations?
Are
they
gonna?
Have
you
know
if
they
don't
have
the
required
skills,
then
you
know
I,
don't
know
who's
gonna
use
it
right,
but
if,
if
it's
not
the
end
JavaScript
developer,
but
somebody
in
the
organization
who's
gonna
have
to
investigate
production
issues,
then
I
still
think
it
makes
sense
to.
You
know:
explain
that
the
the
nodejs
specific
parts
in
there,
which
problems
would
you
use
to
debug?
For,
though,
for
that
target
audience.
B
B
E
Right
so
that,
for
me,
that's
kind
of
important
to
understand
which
use
cases
these
are
important
for,
because
that
leads
to
like
how
important
is
it
to
have
it
enabled
and
available
right,
but
I
mean
not
not
to
affect
anything.
That's
going
on
now,
but
like
longer-term,
it's
kind
of
like
that
would
explain,
would
help
us
to
explain
why,
for
example,
if
there's
a
you
know
a
small
but
noticeable
performance
impact,
you
might
want
to
turn
it
on
versus
not
right.
I
Yeah
I
agree
given
for
couple
of
reasons.
One
is
that
the
the
operating
space
is
thermal,
but
the
problem
determination
space
is
the
JavaScript
in
production.
So
there
is
a
there
is
a
huge
mapping,
that's
required,
so
there's
a
level
of
abstraction,
also
required
number
one
number
two:
that
there
are
a
lot
of
tracing
implementations
when
it
comes
to
the
user
journey
and
note
it's
not
a
very
well
organized
sector
because
you
could
trace
from
the
JavaScript
API
like
trace.
I
Then
system
trace
linux
phrase,
so
many
levels
of
tracing
exists
and
the
exact
lose
cases
or
excite
scenario
where
which
different
kind
of
tracing
is
used
is
not
very
well
understood.
So,
for
these
two
reasons,
I
suggest
or
I
agree
that
we
should
first
run
it
through
the
user
journey
discussion
to
see
which
specific
use
use
cases.
E
G
B
K
B
B
B
B
B
H
Yeah,
and
certainly
on
my
side,
they,
the
kernel
versions
and
getting
some
clarity,
even
if
it's
only
in
my
head
on
that
would
be
great,
so
I
think
those
three
things
or
you're
looking
forward
if
we'd
get
the
bill
from
from
the
unofficial,
from
rod
to
repo
it'd
be
fantastic,
and
certainly
with
the
user
journeys.
I
think
I
mentioned
this
to
you
already
Mateus
and
my
focus
is
on
sort
of
kubernetes
in
the
container
stack,
so
I'll
be
sort
of
trying
to
tie
the
two
of
those
together
as
well.
Yeah.