►
From YouTube: 2022-09-08 meeting
Description
OpenTelemetry Prometheus WG
A
A
A
A
A
C
Working
nice
here,
yeah
I,
live
in
in
Oakland
and
there's
like
a
little
I
guess
they
call
it
a
lake
I,
don't
think
it's
technically
a
lake
but
yeah
there's
like
a
whole
like
algae
bloom,
that's
like
killing
a
bunch
of
fish,
and
so
it's
very
it's
I,
guess
yeah
that
part's
not
normal.
It's
very
Grim
to
walk
by
the
lake
because
it's
like
a
bunch
of
you
know
fish
that
aren't
normally
there
that
have
been
there
recently.
D
Yeah
I
had
on
these
hot
days.
I
had
like
an
outage
like
power
outage
for
several
hours
and
then
like
and
then
like
the
mobile
internet
also
went
down,
and
it
was
like
for
three
hours
like
no
connectivity
and
I
was
like.
Okay
like
this
is
what
it
used
to
look
like
now.
I
need
to
figure
out
what
to
do.
C
D
Yeah
he
has
some
power
like
power,
grid
problems
and,
like
I,
don't
know,
maybe
there's
some
local
Hub
or
switch
or
something
that
also
went,
went
out,
but
yeah
or
I
I
guess
it
also
like
maybe
like
since
the
Wi-Fi
was
out
it
just
like.
So
many
people
started
connecting
to
I
think
it
was
like
intermittent
like
you
couldn't
load
almost
anything
except
like.
If
you
wait
for
five
minutes,
then
maybe
the
web
page
loads.
So
maybe
it
was
just
like
lots
of
strain
on
the
on
the
on
the
network.
Yeah.
A
A
C
See
we'll
give
it
a
couple
minutes.
A
lot
of
people
said
they
weren't
going
to
be
able
to
make
it
today.
I,
don't
know,
maybe
just
being
fall
or
whatever,
but
we'll
see,
I'll
give
like
maybe
a
minute,
but.
A
C
C
C
Most
people
have
been
here
before
I
guess,
but
yeah
welcome
back
everybody.
We
are
I,
guess
almost
there
with
getting
the
original
like
Vision
Otep
that
we
created
approved
and
we're
still
kind
of
sharing
it
with
more
community
members
to
yeah
I
saw
a
tigran.
C
You
posted
in
one
of
the
channels
asking
for
more
people
to
review
comment
all
that
kind
of
stuff,
so
yeah,
if
you
guys
know
anywhere
else
that
it
could
be
shared,
please
feel
free
to
do
so,
so
we
can
kind
of
get
more
eyes
on
it.
B
So
yeah
I
just
wanted
to
say
that
it's
important
that
we
see
a
lot
of
more
approvals
doesn't
have
to
be
actual
official
approver
of
open
Telemetry.
This
group,
for
example,
I,
would
expect
that
I
guess,
let's
say
the
majority
of
the
group
right
at
least
approves
the
the
the
the
the
pr
and
and
if
for
some
reason
you
don't
think
it's
the
right
thing
then
then
speak
up
right
comment
on
it
object.
We
are.
B
We
want
to
see
a
very
broad
alignment
on
this
vision
from
the
community
and
in
particular
from
this
group.
Your
opinion
is
the
most
important
here.
Yes,
as
as
an
official
open,
Telemetry,
approver
I
will
I
will
put
the
check
mark
there,
but
my
voice
is
less
important
than
yours.
Really
all
I'm
saying
is
that
yes,
I
think
it's
useful
for
open
Telemetry,
but
your
voice
is
more
important
because
it
says
yes,
this
is
the
right
approach
from
profiling
perspectives.
So
please
do
review.
Please
comment.
B
A
C
I
will
say:
I
wasn't
totally
clear
on
what
the
process
was
so
I
wasn't
sure
if,
like
only
the
you
know,
sort
of
official
approvers
were
allowed
to
approve
and
I
wouldn't
be
surprised
if
many
others
felt
the
same.
So
I
guess
now
that
we
know
that
we
can
kind
of
share
that
with
people
and
I'm
sure
yeah
some
have
commented
and
then
yeah
a
lot.
A
lot
kind
of
just
gave
it
an
emoji
or
whatever
I.
Imagine
that's
some
sort
of
endorsement,
yes,
sort
of
approval,
but
yeah
foreign.
C
Yeah
and
so
yeah,
so
sharing
that
more
I
guess,
there's
probably
some
channels
in
the
cncf
slack
I'm,
not
sure
which
one
would
be
yeah
I,
don't
know
if,
like
General
is
or
I
mean
I,
don't
know,
yeah
I,
don't
know.
If
there's
one
that
you
might
recommend
there
I.
B
Have
just
posted
in
the
specification
Channel
but
you're
right,
maybe
some
more
channels
like
the
the
General
open,
Telemetry
Channel.
If
there
is
one
I,
don't
remember,
there
should
be
one
probably
yeah.
There
is
one,
maybe
post
there
as
well,
okay,
anywhere,
really
that
you
can
think
of.
There
may
be
people
hanging
out
that
that
may
have
a
an
interest
in
the
topic,
just
just
advertising
right,
so
that
people
know
yeah.
D
And
approval
and
approval
like
if
I
want
to
approve
it.
It's
like,
like
post
an
lgdm
comment
on
DPR
or
is.
B
B
B
B
So
when
you,
when
you
open
the
files
tab,
do
you
not
see
the
the
review
button
when
you
can
just
select
the
approve
and
hit
submit
you?
Don't?
Okay,
so
I
guess
in
that
case
it's
limited
to
the
members.
Only
I.
D
I,
don't
think
so.
Yeah
I
went
to
the
files
tab
and
then
at
the
top
right
corner,
there's
combo
box.
You
can
select
a
proven
click.
Okay,.
A
Okay,
let
me
also
just
write
that
in
case
others
don't
see
it
if
I
was.
C
Happy
to
share
that
around
more
cool
yeah
and
then
the
other.
A
C
Yeah
so
there's
you
know
some
more
steps
out
or
more
steps
that
are
laid
out
in
this
project
management.
Doc
I
will
share.
A
My
screen
to
show
it
two.
C
C
So
I
was
just
kind
of
following
this
I
kind
of
wanted
to
yeah
check
in
with
you
two
gun
and
ask
sort
of
what
you
know
you
think
makes
I
guess
yeah
if
this,
if
there's
anything
missing
from
here
or
if
this
is
actually
a
good
Next
Step
the
you
know,
so
it
says
to
create
a
project
tracking
issue.
C
Are
we
there's
a
issue
for
it
that
yeah
let.
C
And
and
yeah,
and
so
the
the
thing
that
we've
talked
most
about
that
I
kind
of
you
know
would
I
guess
be
a
good
kind
of
seed
for
that
I
I
think
is
the
benchmarking
stuff,
since
that
is
something
that
no
matter
which,
what
route
we
ultimately
go
in
coming
up
with
a
sort
of
more
I
guess:
technical
Otep.
That
will
need
some
sort
of
benchmarking
to
to
measure
you
know.
Whatever
formats
we
eventually
come
up
with
and
and
yeah
we've
already
kind
of
talked
about
it.
C
A
couple
weeks
back,
we
sort
of
put
a
pause
on
it
to
make
sure
we
were
all
aligned
on
the
vision
and
then
so
yeah
I
was
thinking
that
would
make
sense
to
kind
of
seat
it,
but
I
just
wanted
to
kind
of
get
your
thoughts
as
to
you
know,
moving
forward
from
here
what
would
be
sort
of
the
most
I
guess
kind
of
efficient
way
to
move
forward.
B
Yeah,
so
once
we
get
the
the
approvals
there
on
the
old
tab,
I
think
then.
So,
if
we
don't
see
any
objections
like
somebody
finds
something
missing
there,
maybe
hopefully
not
anyway.
So
if
it
gets
approved,
it
gets
enough.
Official
approvals
gets
broad
alignment.
We
see
like
a
large
number
of
approvals
there.
We
can
consider
the
project
to
be
accepted
to
open
climate
right.
So
it's
on
the
board.
B
It's
in
the
it's
in
the
scheduled
projects,
or
whatever
is
the
state
that
we're
seeking
for
then
this
work
group
will
do
whatever
you
consider
is
the
most
important
next
thing
or
first
thing
to
do
right
after
you
have
a
project
as
part
of
open
climate
tree,
if
it
is
the
benchmarking
which
you
do
for
for
the
purpose,
I'm
I'm
guessing
for
the
purpose
of
choosing
the
data
model
and
and
the
format
then
sure
you
can
do
that.
B
If
there
is
something
else
that
you
think
needs
to
be
done,
feel
free
to
do
that
and
you
don't.
You
can
do
that
in
parallel.
Maybe
right,
if
you
need
to
do
multiple
things,
there's
no
need
to
serialize
that
you
have
a
work
group
you
can.
If
you
need
a
repository,
we
can
create
one
for
you.
Then
you
can
start
writing
code.
Do
whatever
you
need
to
do.
Just
do
your
thing
right.
C
Cool
yeah
anybody
else
here
have
any
yeah
and
so
I
guess
yeah.
Just
to
again
that
was
just
kind
of
my
thought
of
what
would
what
would
be
sort
of
like
a
good,
possibly
easy,
first
step
to
start
with
benchmarking,
since
we've
already
done
a
little
there's
already
a
duck,
we've
already
done
a
little
bit
of
talking
about
it
and
we
know
it
will
be
needed
at
some
point.
I
know,
there's
a
lot
of
other
sort
of
you
know,
thoughts
out
there.
C
Obviously,
a
big
one
is
figuring
out
what
the
you
know
final
or
what
a
sort
of
yeah.
What
the
final
data
model
will
be.
I
think
that
one
will
likely
require
a
lot
more
discussion
as
to
you
know
what
fields
we
should
include,
and
you
know
how
we
should
include
them
all
that
kind
of
stuff,
but
yeah
I,
don't
know.
Does
anybody
else
have
any
other
I
guess
ideas
of
you
know
what
what
you
think
would
be
the
next
most
like
pressing
things
to
talk
about?
C
What
would
be
you
know
a
good
kind
of
like
actionable
things
to
work
on
other
than
the
benchmarking
stuff.
D
B
B
That
would
be
my
expectation
right,
an
actual
working
Library,
let's
say
in
go
or
Java
I,
compile
it
with,
and
then
maybe
I
configure
it
somehow
to
choose
whatever
the
sampling
grade
stuff
like
that,
the
destination
to
send
it
to,
and
it
starts
producing
profiling
data
in
the
the
format
that
you
defined
according
to
the
configuration
that
I
defined
when
I
use
your
library.
That
is
the
end
goal
right.
That's
what
we
do
for
traces
for
metrics
and
for
logs
now.
What
are
the
intermediates
that
intermediate
steps
to
to
go?
B
There
I
think
it's
up
to
you
to
decide.
I
I
personally
would
start
by
defining
the
data
model,
the
the
the
maybe
the
data
format
as
a
result
of
that,
and
and
but
it
doesn't
have
to
be
final
and
probably
start
prototyping
experimenting
with
it
to
understand
whether
it's
implementable,
what
are
the
the
performance
characteristic
of
it?
B
If
you
try
to
implement
it,
maybe
in
a
couple
languages
or
maybe
start
with
one
right
depending
on
how
much
time
and
resources
you
have
come
up
with
some
sort
of
a
proof
of
concept,
once
you're
happy
with
it,
then
I
guess
move
forward
right,
start
working
towards
production,
more
production
version
rather
than
just
the
Prototype.
So
maybe
something
like
that
and-
and
you
don't
have
to
feel
obligated
to
follow
any
specific
development
process
here,
guys
really
it's
it's
your
work
group,
your
area
of
expertise.
B
So
you
have
a
lot
of
flexibility
and
freedom
to
choose.
How
exactly
you
want
to
work
here.
We're
really
here
to
support
you
right
with
whatever
you
need.
If
you
need
a
GitHub
Repose,
very
adult
inflammatory,
we
can
provide
it
to
you
or,
if
you
need,
let's
say,
set
up
permissions
and
stuff
like
that
right
help
you
with
SCI
CD,
something
like
that,
then
maybe
we
can
do
that,
but
really
it's
you
that
will
be
doing
the
work
right.
I.
D
B
Yeah,
so
the
faved
ones
are
typically
the
ones
where
people
I
guess
get
bored
or
or
stop
actually
producing
tangible
results,
and
it
goes
into
nowhere
right
as
soon
as
there
is
energy
and
people
are
active
and
moving
towards
that
that
final
goal,
you
should
be
successful.
I,
don't
see
a
reason.
Why
not
because
you
you
have
the
necessary
expertise
in
the
in
the
area
and
we're
here
to
help
you
from
the
perspective
of
being
aligned
without
inflammatory.
D
Yeah
I
think
the
main
kind
of
meta
risk
that
I'm
seeing
is
that
in
the
discussions
of
the
formats
like
there
are
multiple
participants
and
some
participants
might
be
like
secretly
hoping
that
it's
their
format
that
will
be
just
accepted,
as
is
as
kind
of
like
the
standard
for
everyone.
And
then,
if
we
come
up
with
something
that
is
kind
of
like
with
which
technically
addresses
the
requirements
of
everybody.
But
it's
not
exactly
what
they
already
have
and
they
need
to
pay
the
costs
of
migrating
to
that
new
format.
D
Then
people
might
get
like
quietly
disappointed
and
go
away.
That's
kind
of
like
one
of
maybe
I'm
wrong,
but
this
is
one
of
kind
of
like
meta
risks
that
I'm
seeing
yeah.
B
You
you're
totally
right
so
open
Telemetry
is,
is
a
is
a
it's
a
collection
of
people
from
many
in
many
cases
from
competing
companies,
so
there
is
a
risk
out
there.
So
it's
it's
also
an
art
to
to
navigate
that
properly
right,
there's
an
event.
Inevitably,
there
is
politics
involved
in
this
job.
So
be
aware
of
that.
That's
that's
fine
right.
We,
we
managed
so
far
to
find
ways
to
work
together
with
others
from
companies
who
are
competitors
to
ours
and
I
I.
You
should
try
to
do
the
same
right.
B
Otherwise,
it's
not
going
anywhere.
If
everybody
tries
to
pull
it
only
their
way
like
they
benefit
from
it.
Obviously,
there's
not
going
to
be
an
agreement.
So
it's
it's
yeah
again.
It's
it's!
You
need
to
find
the
way
right
so
come
to
an
agreement.
Most
likely.
You
should
aim
for
finding
something
that
unmarried
is
the
best
approach
and
then
convince
others
that
this
is
the
best
approach,
not
because
you
invented
it,
not
because
your
company
uses
it,
but
because
it
really
is
technically
the
best
approach
right.
B
If
you
do
that,
then
you
will
have
a
very
strong
position
to
make
sure
people
agree
with
you.
Otherwise,
if
it's
only
because
the
company
X
does
it
and
the
company
X
is
a
huge
company.
Okay,
maybe
maybe
you
can
try
to
do
that,
but
likely
you're
not
going
very
far
with
that.
We
I
saw
that
it
didn't
work
in
the
past,
so
try
to
do
your
best
as
Engineers
as
designers
to
create
something
that
works
for
the
industry
rather
than
just
for
you
personally
or
for
your
company
yeah.
C
Yeah
I
mean
one
thing:
I'd
kind
of
you
know,
sort
of
add
on
there
is.
That
is
also
part
of
the.
What
I
think
is
a
good.
You
know
motivation
for
finding
some
type
of
objective
metric.
Like
you
know,
you
know,
with
the
benchmarking
Suite
with
something
where
you
can
get.
You
know
this
costs
less
than
that.
This
is
more
efficient
than
that.
You
know
where
you
can
get
something
that's
more
objective,
because
then
it's
you
know
much
harder
to
or
yeah
it's
I
guess
it's
much
easier
to
to.
C
Yeah
like,
like
you
said,
find
something
that
on
Merit
is
the
best
approach
and
people
be
able
to
back
that
up
in
a
engineering
way,
and
so
that's
one
thing
and
then
the
other
thing
is
just
that.
C
I
mean
I
think
there
is
from
when
everybody
did
share
their
respective
formats
and
then
even
when
we
heard
about
you
know
like
P,
Prof
and
JFR
and
stuff
I
mean
one
thing
that
is
kind
of
that
I've
I
think
seemed
promising
was
the
fact
that
not
everything
was
so
different,
so,
like
I,
don't
think
that
there's
like
anything
that
was
like
wildly
different
than
any
one
format
that
was
wildly
different
than
the
other,
but
you
know
there
were
obviously
you
know
little
differences
here
and
there,
and
so
because
of
that
I
do
think
that
we
that
there
is
a
path
to
you
know
to
an
agreement.
D
D
I
feel
like
aggregated
profile
versus
kind
of
like
almost
Trace
profile,
that's
one
because
I
think
some
people
assume
and
and
want
the
timestamps
and
I'm
not
sure
like
we
answered
that
completely
another
one
is
incrementality
and
statefulness
of
the
protocol,
because
some
people
wanted
that
and
how
much
the
how
much
the
resulting
format
will
support
that.
That's
an
open
question
for
me
and
and
third
is
meta.
D
Self-Descript
descriptiveness
of
the
format
because
it
seems
like
GFR,
is
able
of
essentially
like
carrying
proto-descriptors
inside
of
the
format
itself,
and
also
some
people
mentioned
that
like
well.
We
like
GFR,
because
it's
like
it's
kind
of
like
you
extended.
You
can
extend
it
with
arbitrary
data
and
the
data
is
described
in
the
format
itself.
So
it's
this
guy,
like
it's
so
Universal
and
I,
don't
think
we
will
need
to
put
a
steak
somewhere
and
say
like
well,
we'll
not
have
this.
We
will.
We
will
have
a
Proto.
We,
the
product,
contains
what
it
contains.
D
We
will
not
have
a
like
proto-descriptors
inside
of
this
product,
so
those
are
three
things
that
and
I
I
agree
with
you
that
benchmarking
will
likely
help
to
answer
many
of
this
as
long
as
we
say
as
I
think
but
I
think
it
will
with
benchmarks.
It
will
be
important
to
say,
like
here's,
the
set
of
Benchmark
data
and
Benchmark
formats
that
we
got
that
represents
the
requirements.
Does
everyone
agree
because
otherwise
people
might
silently
assume?
D
Well,
we
don't
have
yet
benchmarking
data
with
timestamps,
but
probably
it's
coming
so
I
will
wait
and
hopefully
we
will
support
it.
Like
I
think
we
will
need
to
be
very
explicit
that
we
think
this
pinch
marking
data
set
now
represents
the
requirements
that
we
want
to
address.
Does
everyone
agree
something
like
that.
C
Yeah,
that
makes
sense.
Can
you
can
you
explain
more
what
you
meant
when
you
said
time
stamps
what
the.
D
Like
Sample
timestamps
versus,
for
example,
if
you
try
to
put
timestamp
data
into
into
the
profile
Proto
that
will
not
be
very
optimal
today,
because
you
can
put
a
timestamp
as
a
label,
but
there
is
no
like
special
unit
for
timestamp
label.
You
will
have
to
duplicate
the
stacks
like
do.
We
need
that
in
the
profiler
that
I
worked
on,
we
didn't
have
those
like
timestamps
for
individual
samples.
We
usually
have
like
10
second
collected
data,
and
it's
aggregated.
D
There
are
no
timestamps,
except
like
timestamp
for
the
whole
profile
that
this
profile
was
collected
for
10
seconds
at
this
point.
At
this
point
in
time,
and
I
got
the
impression
that
some
people
wanted
to
actually
have
like
individual
samples
within
the
collected
data.
Does
it
does
it
do
I?
Explain
it
better
or
it's
so
confusing.
B
So
those
are
great
topics
to
to
debate
about
right.
So
you
listed
three
things.
Maybe,
though,
each
one
of
those
is
a
topic
of
a
discussion
and
then
if
there
is
a
disagreement
about
how
to
move
forward
with
that,
then
you
show
your
solution.
The
benefits
of
your
solution.
Right,
you
show
a
prototype.
You
show
why
it
works
better
than
the
the
opposite
approach
right
and
that's
that's
how
you
win
right.
I
I
wanted
to
say
one
more
thing,
guys
on
the
formats
data
model.
B
We
we
have
seen
that
you
do
not
necessarily
have
to
choose
something
that
is
already
widely
accepted
in
the
industry,
with
with
open
Telemetry
protocol,
for
example,
we
it
was
a.
It
was
a
completely
from
scratch
design
right.
It
did
not
try
to
follow
any
specific
existing
approach,
yet
it
proved
to
be
successful.
B
So
there
is
enough
momentum
behind
open
Telemetry
that
you
now
have
the
opportunity
to
come
up
with
something
that
is
different
from
everything
that
exists
doesn't
have
to
be,
but
it
can
be
and
still
be
successful
and
still
be
supported
by
by
the
vendors
in
the
industry.
We
when
we
introduced
otlp,
it
was
nothing
right
in
three
years,
virtually
every
vendor
in
the
industry
supports
it
today.
So
you
can.
You
can
I,
guess
a
few
Freer
here
not
to
be
constrained
by
the
the
past
of
the
profiling.
B
Let's
say
something
is
not
working
well,
but
it
is
the
de
facto
standard.
Today
you
have
the
opportunity
to
improve,
to
to
I
guess
to
to
some
somehow
be
free
from
that
Legacy,
if
necessary.
So
I
would
encourage
you
to
feel
fear
to
design
the
best
approach
that
you
can
come
up
with
and
then
let
the
people
in
the
industry
worry
about
how
they
they
support
him
in
their
backend.
Obviously,
your
design
needs
to
be
supportable
right.
B
B
C
See
one
question
I
kind
of
had
was
for
like
tracing
and
I.
Don't
know,
maybe
there's
a
better
person
to
ask,
but
did
did
Like,
Jaeger
and
I.
Guess
yeah
like
what
Tracer
sort
of
preceded
the
like
open,
Telemetry
tracing
efforts
and
how
was
it
I
guess
like
dealt
with
to
make
sure
that
they
were
I?
Guess
you
know
yeah
that
people
who
were
already
using
the
if
it
was
before
I
guess
yeah.
First
of
all
did
the
did
those
tracers
kind
of
proceed.
The
open,
Telemetry
efforts
so.
B
Open
sensors
and
open
tracing
ideas,
it
could
interrupt
inflammatory,
okay,
the
open
Telemetry
protocols,
tracing
data,
modeling
format,
I
designed
it,
but
when
designing
it,
I
looked
into
Jagger
Zipkin,
open
sensors.
A
couple
others
and
I
made
sure
that
the
design
made
it
possible
to
represent
all
of
these
formats
properly
right
so
that
we
could
do
the
translation,
unambiguous
translation
from
this
formats
into
otlp
and
back,
if
necessary,
and
and
and
with
actual
implementations
of
that,
and
with
benchmarking.
That
compares
how
the
proposal
was
better
or
no
worse.
A
See
I
believe.
B
It
could
be
yes,
I
can
share
it
again.
If
you
want.
A
Yeah,
that
would
be
I'll
put
it
in
the
notes.
That
would
probably
be
useful
to
to
see
okay,
yeah,
yeah
I.
Think
that
makes
sense
paste
this
in
the
notes.
C
So
yeah
I
think
all
those
are
yeah
I
think
all
those
are
good
topics,
the
ones
that
you
mentioned.
Alexa
and
yeah
I,
definitely
think
if
I
remember
right,
yeah
the
statefulness
thing
was
pretty
yeah.
I
would
say
a
pretty
big
topic.
C
Yeah
I
agree:
okay,.
E
And
the
statefulness
was
potentially
also
something
that
you
know
we
could
make.
You
know
Implement
in
a
way
that
you
could
use
it.
If
you
want
to
and
not
use
it,
if
you
don't,
the
extendability
of
the
format,
I
think
is
the
biggest
one
where
you
can
only
really
go
one
way
or
the
other
I
suppose.
E
E
B
D
Yeah
yeah,
of
course,
like
backward
compatibility
and
yeah.
Oh
yeah
I
only
meant
like
no
I,
think
product
within
the
product,
yeah
and
and
even
even
that,
like
to
some
limited.
It's
also
like
it's,
for
example,
profile
Proto
has
like
has
like
comments
field
and
technically
you
could
put
anything
into
that
common
field
like
you
could
put
Json,
but
we
will
assume
good
intention,
yeah
people
in
it
yeah.
As
soon
as
you
have
a
string
field,
all
kinds
of
interesting
things
can
happen.
A
C
All
right,
cool,
I
think
all
that
makes
sense.
You
know
yeah
I
think,
especially
once
we
you
know
at
Future
meetings,
we
can
plan
to
devote
kind
of
dedicated
time
to
discussing
the
more
hairy
topics
there
yeah
in
the
meantime,
I
thought
it
might
be
useful
to
go
back
over
the
benchmarking
dock.
C
It
might
be
I'm
curious
what
you
guys
think
we
could
yeah
I,
don't
know,
I
think
it
would
be
useful
to
kind
of
like
sort
of
recap
what
was
in
there
at
least,
and
maybe
we
can
see
if
that
will
help
with
some
of
the
evaluation
and
some
of
the
discussions
in
the
future,
and
if
it's
worth
you
know,
if
there's
enough
agreement,
even
though
yeah
we
have
a
smaller
group
than
we
normally
have
today.
C
If
there's
enough
agreement
to
sort
of
Warrant,
you
know
yeah
like
creating
a
repo
and
sort
of
getting
started
on
that
process
in
parallel.
What
do
you
any
I
guess
any
objections
to
that.
C
So
Dimitri
I
know
you
did
a
lot
of
the
I
guess
the
bulk
of
the
the
writing
here.
Do
you
want
to
kind
of
walk
through
sort
of
the
the
thought
process
of
the
duck.
C
E
Was
just
saying
that
it's
the
same
dog
that
we
discussed
before
and
we
just
kind
of
polished
it
a
little
more
the
idea,
the
basic
idea
is
still
the
same.
E
It
is
to
have
a
bunch
of
test
cases
of
profiles
of
you
know
different
kinds,
maybe
profiles
that
are
very
that
have
a
lot
of
different
symbol,
names
or
profiles
that
are
very
big
trees
and
the
opposite.
You
know,
have
a
diverse
variety
of
test
cases
and
then
to
have
a
reference
implementation
for
the
encoder
that
would
take
these
profiles.
E
Encode
them
into
you
know
the
the
format
there
will
eventually
become
more
open,
Telemetry
format
and
then
measure
how
much
time
it
took
and
what
is
the
resulting
size
of
the
profiles,
and
hopefully
these
metrics
will
help
us,
as,
as
everyone
was
saying,
to
kind
of
make
decisions
in
a
more.
You
know,
objective
way.
I
suppose
should.
E
E
If
we
can
yeah
I,
don't
see
why
not
I
think
like
yeah
I,
think
the
the
two
measurements
that
we
absolutely
have
to
make
are
the
you
know
the
the
size
and
the
amount
of
CPU
it
takes.
E
I
guess,
yeah
memory
is
also
one
of
one
of
the
important
ones.
One
thing
I'm
slightly
concerned
about
is
that
I.
Imagine
it's
harder
to
measure
memory
usage,
but
I
might
be
wrong.
There.
D
I
think
go
benchmarking
can
do
that
I
think
it
can
like
I,
don't
think
it's
like
Peak
memory
usage,
but
it
was
like
stats
on
total
allegations
or
something
I
would
yeah.
A
Be
the
memory
usage
one
is
very
tricky
because
you'd
have
to
have
the
input
the
profiles
be
held
in
whatever
in-memory
formats.
The
profile
actually
generates
them
in
because
potentially
the
encoder's
just
going
to
be
doing
pointer
copies,
which
is
very
different
to
it,
having
to
copy
the
data
itself
right.
A
So
if,
if
your
input,
if
your
profile
is
in
some
synthetic
format,
most
likely,
you
know
rather
than
text
file
or
something
like
that,
your
memory
usage
is
going
to
look
very
different.
E
Okay,
cool
I,
guess
maybe
when
and
when
we
like
I'm
thinking.
What
are
we
going
to
compare
we're,
probably
going
to
compare
different
runs
of
this
reference
implementation
right?
Maybe
someone
will
make
make
changes
to
the
reference
implementation
and
then
compare
kind
of
like
before
and
after
so
in
that
case,
maybe
the
effects
of
of
this
thing,
you're
talking
about
Jonathan,
will
be
kind
of
you
know
diminished,
but
maybe
maybe
I'm
missing.
Something
like
you
see
what
I
mean
like
when
you
compare
the
two
things.
C
Yeah
I
think
that
makes
sense
at
the
very
least
yeah
we
can
yeah,
you
know,
create
an
issue
for
it
and
if
yeah.
E
E
Yeah
I
think
yeah.
We
could.
We
could
start
with.
You
know
a
small
set
of
profiles,
small
set
of
metrics,
very
simple,
encoder
and
kind
of
go
from
there.
It
will
be
easier
to
to
improve
a
very
simple
encoder
too.
B
Yeah
so
I
guess
you
should
also
compare
to
some
well-known
formats,
or
maybe
one
format,
at
least
right,
to
show
that
what
you're
proposing
is
in
some
ways
better
than
what
already
exists.
Otherwise,
why
are
you
proposing
it
right?
Yeah.
A
E
Yeah,
we
were
also
yeah
thinking
about
encoding
it
as
existing
hotel
events.
I
think
that
was
another
idea
to
see
if
you
know
that
that
you
know
if
we
could
use
that.
C
Yeah
I
think
you
had
mentioned
that
a
while
back
to
you
on
that.
You
know
it
might
be
a
good
sort
of
like
Baseline
to
ensure
that
it
at
the
very
least,
is
you
know
worth
an
entirely
new
thing:
yeah
compared
to
yeah
yeah.
B
A
D
Exactly
that
that
means
like
use
existing
because
there's
no
like
exact,
open,
Telemetry
profiling
format,
but
it
doesn't
mean,
like
put
I,
don't
know
like
the
stack
of
the
sample
separated
with
slashers
as
a
log
line
and
then
and
then
person
they're
still
going
to
be
like
a
little
format
within
the
format.
So.
B
Yes,
exactly
so
open
Telemetry
log
records,
They
are
structured
data,
I
think
equivalent
of
Json
right.
So
you
can
put
anything
there.
If
you
want
right,
you
you
can
put
numbers
I,
don't
know
strings,
key
value,
lists
arrays
and
all
that
stuff
is
representable
there.
So
that
is
generic
enough,
that
you
can
I,
guess
dissect
your
profile
in
a
way
that
that
you
put
it
there
right
so
shouldn't
be
very
difficult.
To
do
likely
is
not
going
to
be
very
optimal,
but
it's
it's
not
a
bad
thing
again.
B
E
And
yeah,
another
part
of
this
proposal
is
designing
and
I
guess
like
building
some
sort
of
a
program
that
would
take
profiles
in
JFR,
P,
Prof
and
other
popular
formats
and
and
anonymize
data,
so
that
people
could
contribute
profiles
without
you
know
risking
any
sort
of
privacy,
things
yeah
I
kind
of
wonder
what
you
guys
think
about
that
one
like
how
how
important
it
is
to
have
it.
Maybe
how
important
it
is
to
have
it
early.
C
I
think
it
would
be
interesting
honestly,
you
know,
honestly,
we
could,
theoretically
kind
of
you
know,
find
a
way
to
get
a
bunch
of
different
profiles
from
a
bunch
of
different
languages
that
have
different.
You
know.
Characteristics
of
you
know
how
big
the
stacks
are,
and
all
that
kind
of
stuff,
but
I
think
it
would
be
good
from
a
you
know.
A
potential
way
to
like
get
people
in
the
community
also
somewhat
bought
into
it.
C
If
they're,
you
know
able
to
say
that
they
contributed
the
types
of
stacks
that
their
organization
or
whatever
that
they
that
they
have
or
the
type
of
profiles,
I
guess
I
think
it
would
be
interesting
from
that
standpoint.
I,
don't
know
how
I
guess
how
important
it
is.
C
You
know
outside
of
that,
because
then
I
guess
yeah.
We
could
obviously
just
synthetically,
make
a
bunch
of
profiles
and
and
do
that,
but
I
don't
know,
I
think
it
would
be
an
interesting.
C
You
know
interesting
to
see
how
much
you
know
input
people
get
who
would
yeah
like
submit
PRS
with
profiles
to
to
put
into
this
giant.
You
know
I.
B
Guess
yeah,
so
so
there
is
one
more
option:
there
I
think
it's
it's
probably
useful
to
try
to
come
up
with
real
production
data
and
anonymize
it
and
then
run
your
benchmarks
on
it,
but
it
may
not
be
easy
to
obtain
and
it
depending
on
who
obtains
it
from
where
you
may
get
different
results.
So
as
an
alternate
to
that
and
maybe
to
avoid
creating
this
anonymizer,
you
could
try
the
open,
Telemetry
demo
application.
B
We
have
a
demo
which
includes
services
in
multiple
languages,
and
maybe
you
could
try
to
see
what
do
you
generate
from
from
this
demo
right
in
the
language
that
you
want
to
test.
There
is,
let
me
send
the
link
to
that
demo.
It's
a
new
one
hold
on
a
second
yeah,
so
you
could
you
could
make
this
demo
generate
profiles
right,
and
that
could
be
that
your
reference
data
on
which
you
you
run
the
the
tests.
A
B
B
The
the
the
there
is
some
description
about
how
you
can
generate
traffic
towards
it.
I
didn't
try
it,
but
that
was
the
intent.
So
this
is
the
back
end
composed
of
multiple
services
in
different
languages
and
here's
how
you
can
generate
some
traffic
to
it
to
start
seeing
actual
Telemetry
coming
out
of
it.
Yeah.
C
Yeah
so
yeah,
that's
actually
maybe
that's
a
good
place
to
start,
and
then
you
know
we
can
kind
of
revisit.
Do
it
trying
it
with
real
data
if
we
feel
like
that
would
add
value
to
it
after
yeah.
D
Also
I
know
we
started
discussions
on
this
I.
Don't
I,
don't
know
if
you
finish
them,
but
I
think
one
important
thing
is
to
agree
on
the
intermediate
format,
because,
because
some
of
the
risks
that
I
listed
earlier
in
some
of
the
potential
contentious
points,
those
three
ones
I
think
they
might
bubble
up
during
discussing
and
finalizing
the
intermediate
format,
because
intermediate
format
is
kind
of
it's
almost
like
a
mini
version
of
the
capabilities
of
the
future
proposal.
D
E
D
I
think
we
wanted
to
I
think
we
wanted
to
avoid
like
M
by
m
problem.
So
one
thing
is:
intermediate:
format
is
only
for
the
benchmarking.
It's
not
any.
It's
not
any
part
of
the
future
proposal.
It's
simply
take
a
bunch
of
formats
in
Industry
formats,
convert
them
to
the
same
intermediate
format,
and
then
let
people
write
converters
from
intermediate
format
to
their
proposal
proposed
formats
so
that
we
can
compare
them.
But
we
don't
need
to
write
like
GFR
to
proposal
one
people
of
two
proposal,
one
GFR
to
proposal
two.
Yes,.
A
Is
it
right
that
the
intermediate
format
is
not
part
of
the
the
final
spec,
because
thinking
in
terms
of
what
you
might
want
to
do
in
the
in
the
collect
or
in
the
client
you're
going
to
need
some
kind
of
API
level
representation
of
the
data?
In
order
that
you
can,
for
example,
filter
it
or
anonymize
it
before
it's
encoded?.
A
A
C
Was
just
going
to
say
that
I
guess
some
of
the
feedback
initially
was
that
or
the
ideas
initially
was
that
this
intermediate
format
is
sort
of,
like
kind
of
as
as
like
I
I
guess
like
raw
isn't
the
best
word
but
like
as
simple
and
raw
as
it
could
be,
so
that
it
is,
as
you
know,
sort
of
like
everything
you
know
sort
of
yeah.
It's
like
the
common
denominator,
and
so
we
had
talked
about
I.
Think
initially,
this
doc
proposed,
like
P
profits,
the
intermediate
format
and
people
were
like.
C
E
Yeah
and
the
concern
was
that
with
P
Prof
like
we,
we
in
this
Benchmark,
we
might
be
really
benchmarking.
You
know
the
conversion
process
from
P
Prof.
To
this
you
know
new
format,
which
is
not
really
what's
going
to
happen.
In
most
cases,
what's
going
to
happen
in
most
cases
is
profiler
will
have
data
coming
in
in
some
format,
and
we
need
to
encode
that
right
and
what
is
the
closest
to
that?
You
know
profiler,
and
we
decided
that
you
know
like
some
sort
of
a
text
format.
E
D
E
D
B
So
what
you
need
to
Benchmark
is
how
much
time
it
takes
for
you
to
serialize
the
data
in
memory,
the
profiling
data
into
the
wire
format
that
you're
proposing
and
the
opposite
direction
from
CPU
usage
and
memory
usage
perspective
and
from
the
serialized
data
size
perspective
right,
but
not
the
time
it
takes
to
convert
to
this
intermediate
format
or
back
from
it.
That
should
be
excluded.
D
Maybe
there
should
be
some
kind
of
like
two
explicit
stages
for
each
Benchmark.
One
is
like
read
and
another
is
like
rights
and
once
we
like,
we
don't
measure
the
read
the
read
phase,
but
we
might
measure
total
memory
consumption
after
the
read
phase,
because
that
will
represent
like
what
is
the
cost
of
holding
the
memory
in
this
format.
And
then
we
measure,
and
then
we
measure
the
right
phase
and
for
that
we
could
measure
like
IO,
bytes,
CPU
time
spent.
B
E
E
Yeah
and
so
for
the
format
itself,
the
proposed
format
is
basically
the
kind
of
the
commonly
used
collapse.
Format
where
you
have
stack
traces,
limited
by
semicolons
and
with
number
of
samples
at
the
end
of
the
line,
and
it
is
extended
with
these
two
things
which
are
labels,
they're,
comma,
separated
with
you
know
this
kind
of
format,
and
then
the
other
thing
is
timestamp
that
is
and
and
both
of
these
are
optional,
so
you
know
you
can
you
can
have
them?
You
cannot
have
them,
but.
B
C
So
yeah
I
guess
we
have
you
know
a
couple
more
minutes,
I
mean
yeah.
Do
you
all
think
that
this
is
a
you
know
reasonable
enough
kind
of
like
path
that
we
could
start?
C
You
know
actually
like
turning
this
into.
You
know
issues
getting
a
repo
to
like
you
know,
hold
all
of
this
stuff.
C
You
know
that
kind
of
thing,
creating
a
project
board
for
it
that
kind
of
stuff
yeah
I'm
curious.
If
you
you
think
or
if
you
think
we
should,
you
know,
discuss
more
before
taking
those
steps.
D
I
think
it's
reasonable
I
would
probably
bring
some
of
the
like
details.
What
we
discussed
today
to
the
next
meeting
to
discuss
with
more
people
but
I,
don't
think
they
should
be
blocking.
C
Cool
yeah
I
mean
I,
think
yeah
and
I
was
thinking.
You
know
if
yeah
I
mean
it
seems
like
something
that
could
be
sort
of
worked
on
in
parallel.
C
If
and
we
will
definitely
yeah,
obviously
discuss
more.
You
know
with
everybody,
you
know
at
Future
meetings,
but
it
did
seem
like
something
that
would
be
nice
to
get
the
ball
moving
in
parallel,
because
I'm
sure
there's
just
going
to
be
a
lot
of
kind
of
yeah
figuring
out
yeah
like
making
the
repo
setting
up
the
project
board,
organizing
it
and
I
thought
that,
would
that
would
make
sense
to
do
and
then
maybe
we
can
kind
of
like
have
that
part?
C
You
know
somewhat
ready
by
the
next
meeting,
if
not
fully
ready,
and
then
we
can
kind
of
adjust
that
as
we
discuss
and
yeah
and
kind
of
have
a
good
get
into
a
good
sort
of
workflow
there
of
using.
You
know,
GitHub
issues
and
actually
I'm
getting
some
code
written.
C
B
There's
any
if
there
is
any
concern
that
the
choice
of
the
format
somehow
can
Heaven,
if
you
can,
let's
say,
choose
a
format
in
a
way
that
is
more
suitable
for
a
particular
language
and
less
suitable
for
another
one.
Then
you
would
probably
want
to
demonstrate
that
your
format
is
overall,
most
suitable
for
a
variety
of
languages.
Right
I,
don't
know
if
that's
the
case,
but
if
it
is
then
showing
how
it
works
in
just
one
language
may
not
be
sufficient.
A
D
B
There's
like
10
more
than
10
languages
in
hotel,
supported
today,
but
I.
Think
probably
you
should
I
guess.
If
you
were
asking
for
the
most
popular
ones,
it
would
be
maybe
Java
python
goal.
Maybe
yes,
three
ogs
not
GS,
maybe
yeah
those
four
I.
Guess
it's
probably
too
much
to
do
all
four
in
the
Prototype
again,
maybe
in
the
Prototype,
you
just
start
with
one,
but
if
there
are
concerns
about
the
suitability
of
the
format
for
a
particular
language
or
class
of
languages,
I
don't
know
in
inter-credit
languages.
B
D
C
But
yeah,
that's
a
detailey,
yeah
I
mean
I.
I.
Do
think
that
I
would
suspect
that
if
everybody
picked
two
languages,
you
know
that
maybe
not
everybody,
but
that
the
majority
would
have
go
as
one
of
those
languages.
So
maybe
I
don't
know
yeah.
That
might
be
a
good
place
to
start,
but
yeah.
Well,
all
yeah
post.
C
These
notes
in
the
channel
and
yeah
ask
you
know
yeah
see
if
anybody
has
any
thoughts
on
that
and
otherwise
yeah
I
think
we
can
kind
of
move
a
little
bit
forward
here
and
then
we'll
discuss
all
this
plus
the
the
main
sort
of
points
of
contention
that
you
brought
up
Alexa
at
the
next
meeting.
C
Good
cool
any
other
thoughts
before
we
hop
off
here
from
anybody
all
right.
Well,
thanks
everybody
and
we'll
see
you
all
in
two
weeks.