►
From YouTube: 2022-08-02 meeting
Description
cncf-opentelemetry meeting-2's Personal Meeting Room
A
A
B
A
Kind
of
tired,
so
I
might
not
be
as
animated
I
might
just
harangue
for
eating
stuff.
That's
apparently
my
mode
this
morning,
so
excellent,
so
careful
what
you
put
on
camera,
matt,
yeah,
there's
there
will
be
no
eating
over
here.
Drinking
too
see.
I
have
one
sip
of
coffee
left,
wow,
very
cool,
very
cool.
B
B
What
were
you
using
to
make
coffee
before
we
had
a
french
press
and
we
had
a
whatever?
The
glass
thing
is,
where
you
put
the.
B
Both
of
those
methods
yeah,
we
had,
we
were
living
overseas.
We
got
a
nespresso
they're
like
really
cheap
over
there
because
they're
not
like
imported
or
whatever
like
so
we
had
an
espresso,
but
then
over
here
the
capsules
are
incredibly
expensive.
So
I
looked
at
the
end
of
the
month.
One
month
my
partner
is
just
like:
hey,
like
we
spent
all
our
money
on
this
espresso
capsule,
so.
B
A
There
was
a
there's,
a
discussion.
There
are
these
people
emily
and
bruno
they
work
for
the
same
company.
I
feel
what
company
that
is
up
here.
A
Who
knows,
but
they
were
trying
to
standardize
hotel,
sdk
enabled
as
an
environment
variable.
Apparently
it
comes
from
the
either
the
java
sdk
or
the
java
java,
auto
instrumentation.
A
Library
and
people
seemed
quite
unhappy
with
this
change.
Actually,
so
I
think
there's
a
lot
of
questions
to
ask
why
this
is
even
necessary
and
why
you
wouldn't
just
kind
of
set
a
handful
of
things
to
be
no
op
implementations
yourself,
but
I
guess
we'll
see
where.
D
A
A
We
have
talked
about
this
several
weeks
ongoing.
It's
the
partial
success
changes
to
otop,
and
it
always
makes
the
most
sense
to
me.
If
I
look
at
the
proto,
but
basically
when
you
send
a
a
post
of
traces
or
metrics,
you
can
indicate
how
many
it
used
to
be
yeah.
The
change
here
was
that
you
had
a
number
of
accepted
records.
D
A
To
say
the
number
of
rejected
records,
and
I
think
that
that
makes
it
easier
to
know
what
to
do
if
you
don't
get
a
value
or
if
the
default
value
ends
up
getting
kind
of
sent
back,
whereas
if
you
had
successful
records
and
the
default
value
was
a
zero,
it
would
kind
of
look
like
everything
failed.
But
this
thing
it
actually
kind
of
works
out.
If
you
just
have
these
on
set
and
receive
like
a
people.
A
Has
a
a
namespace-
and
I
think
it's
an
attribute
that
attempts
to
avoid
conflicting
metric
names
between
different
instrumentation
libraries.
A
A
Yeah,
I
think
one
of
the
discussions
here
was
how
this
should
work.
If
you
are
exporting
to
certain
systems
like,
should
this
actually
be
prefixed
on
to
a
metric
name
or
not?
A
And
there's
oh
man,
yeah
j
mcd
was
just
calling
out
that
there
are
a
handful
of
exponential
histogram
related.
A
All
right
what
else
happened?
Okay,
the
metrics
1.0
progress
report.
A
It
seems
like
hotel
java,
hotel,
python
and
hotelnet
are
apparently
stable.
I
feel
like
I'm
not
totally
sure.
If
this
is
true,
I
feel
like
hotel.
Python
might
still
be
like
approaching
a
ga,
but
I
could
be
wrong.
There.
Hotel
go
and
hotel
js
are
soon
to
have
rc's
released.
A
And
then
the
last
question
was
around
this
metric
type.
This
summary
metric
type,
these
are
widely
used
in
prometheus.
They
exist
in
in
otlp,
there's
a
there's,
a
definition
of
a
summary
point,
but
there's
no
like
summary
in
the
api,
so
this
person
was
kind
of
asking.
How
can
they
output
summaries?
A
A
Or
the
answer
is
that
what
they
would
like
to
be
possible
is
that
summaries
can
be
derived
from
histograms.
If
you
have
like
a
summary
aggregation,
somehow
hooked
up
to
your
instagram,
whatever
convoluted
thing
you
have
to
go
through
to
actually
make
that
happen,
it
seems
like
in
theory
that
should
work
as
if
they
can
practice.
A
It's
unclear
how
well
any
of
the
sdks
or
apis
or
interfaces
are
set
up
to
make
this
happen,
but
I
think
that's
what
they
would
like
to
happen
and
there
were
rumors
that
maybe
java
has
this
capability
already
somehow
and
that's
what
this
person
was
using
was
java.
So.
B
A
All
right
so
so
yeah,
let's
take
a
look
at
our
agenda.
B
I
I
have
one
I
forgot
to
add
it
I'll.
Do
it
right
now
it's
about
the
race
car
instrumentation,
which
actually
arielle
commented
on
a
bit
before
this
meeting,
but.
B
Sorry,
I'm
finding
the
pr
I
had
one
knit:
oh
okay,
yeah,
so
let
me
drop
it
in
the
chat.
Here's,
the
pr
and
the
user
chris
holmes
who's
been
a
contributor
before
had
a
bunch
of
good
feedback,
has
been
working
on
this
pretty
diligently
and
they
basically
are
just
last
week
we
talked
about,
they
had
implemented
some
of
their
instrumentation
by
monkey
patching
and
we
sort
of
went
back
and
forth
on.
B
It
was
possible
to
leverage
asn
active
support
notifications
to
do
that
which
is
good,
and
so
we
suggested
them
using
export
notifications
if
they
could,
which
they
can.
They
still
need
to
do
some
monkey
patching
for
context,
propagation
and
some
of
the
for
batch
processing.
B
It's
not
possible
to
use
active
support
notifications,
and
so
for
now
they
just
removed
the
batch
processing,
instrumentation
actually
and
just
said
they
wanted
to
work
upstream
with
the
race
car
maintainers
to
get
asn
stuff
added.
But
in
the
meantime
they
just
said.
Let
me
just
not
instrument
it,
but
the
anyway,
so
they
added
active
support
notifications.
The
one
thing
I
thought
that
was
like
a
little
unholy
that
I
wasn't
sure
what
like
the
right
approach
is.
B
Is
they
made
active
support,
a
explicit
dependency
of
the
gem
like
in
the
gem
spec
it
the
the
gem
spec
gem
gemfile?
Sorry,
I'm
talking
and
trying
to
see
what
you
just
control
f
for
gem
spec,
so
yeah.
B
I
wasn't
sure
in
my
mind
that
shouldn't
be
I
kind
of
looked
at
like
what
the
precedent
is
for
other
instrumentation
that
relies
on
action,
support
notifications,
so
active
model
serializers
uses
it
and
they
just
do
a
compatibility
check
for
like
the
class
or
like
the
module
or
whatever
right
for
you
know
like
whatever
active
support
notification
classes
and
if
it
doesn't
exist,
then
it
fails.
The
compatibility
check,
which
I
thought
maybe
was
the
right.
I
can
link
to
that.
So
I
wasn't
sure
whether
that
was
valid
feedback.
B
I
think
it
is,
but
I'm
not
sure-
and
I
kind
of
wanted
other
folks
opinion.
The
other
question
is
they're,
just
using
like
raw
actual
access
point
notifications,
they're
not
using
like
our
instrumentation
gem,
like
the
active
support
instrumentation,
which
we
could
also
discuss
so
yeah.
Those
are
my
two
things.
If
anyone
is
familiar
with
those.
A
So
yeah,
I
think,
ideally
we
would
want
us
to
be
able
to
be
part
of
instrumentation
all
and
the
end
user
users
should
not
have
to
have
action.
Support
notifications.
If
they're
you
know
running
outside
of
a
rails,
environment.
A
Probably
in
like
the
compatible
check
or
something
you
would
just
see
if
active
support,
the
constant
exists
or
something
along
those
lines
and
log
some
sort
of
warning,
like
you
know,
we
could.
B
The
other
was
right
now,
it's
it
sounded
like.
Actually
I'm
looking
into
it
too
closely.
It
sounded
like
there
was
some
features.
They
wanted
from
active
support
notifications
that,
like
some
configurability,
that
they
couldn't
get
by
using
the
active
support
notification,
instrumentation
gem
arielle,
do
you?
Are
you
familiar
with
that.
C
Taking
a
couple
of
steps
back
so
I
I
want
to
go
back
to
go
back
to
what
you
had
said.
It
is
my
understanding
and
I
think
that
we
should
socialize
this.
If
it's
not
the
case,
instrumentation
should
only
depend
on
other
instrumentations
or
the
the
registry
or
the
hotel
api
and
should
not
have
a
dependency
on
an
sdk,
specific
implementation
and
should
not
have
runtime
dependencies
on
third-party
libraries.
C
They
should
only
be
development
dependencies
and
I
don't
know
specifically,
I
haven't
looked.
I
didn't
notice
a
comment
saying
that
the
usage
of
our
rapper
or
you
know
our
hotel
active
support.
C
Implementation
like
what
hours
was
missing.
I
didn't
see
that
comment.
Did
he
comment
after
I
made
a.
C
C
If
we
add
a
configuration
object
for
that,
I'm
happy
to
swap
it
over
the
span.
Name
would
end
up
being
the
message
you
subscribe
to
and
it
wouldn't
be
the
semantic
name
of
the
messaging
component,
which
would
be
message
received,
and
so,
in
addition
to
transforming
the
payload,
I
think
chris
is
saying
here
that
there's
a
missing
option
to
customize
the
span.
B
Fair
enough
seems
straightforward
to
add
his
configuration.
D
Yeah,
I
know
I'm
thinking,
I'm
thinking,
because
I
honestly
I
got
I
had
gotten
tim
to
do
a
lot
of
the
active
support
stuff.
So
I'm
trying
to
remember
the
race
car
was
set
up
was
very
similar
to
the
ruby
kafka
instrumentation
and
I
don't
believe
we
have
any
like
support
hooks
for
that.
B
This
is,
I
think,
chris
also
authored
the
arty
kafka
instrumentation,
so
he
probably
is
good
contacts.
I'm
talking
as
I'm
I'm
just
trying
to
distract
everyone.
When
I
actually
read
the
instrumentation.
C
C
Faraday's
fans
with
something
so
the
related
conversation
that
I'm
having
right
now
with
some
of
my
teammates
is,
we
want
to
add
auto
instrumentation
for
the
twerp
library.
C
C
But
right
now
that
you
know
shared
context,
only
shares
attributes.
It
doesn't
do
anything
else
or
give
you
access
to
the
the
child
span
in
order
for
you
to
make
any
further
modifications
to
it,
to
enhance
it
or
to
provide
a
little
bit
more
color.
D
Yeah,
it's
very
much
like
meant
to
it's.
Just
like
it's
kind
of
a
one-way
street
like
here's,
some
stuff
that
you
can
tack
on
later,
whereas
like
the
rackspan
hook,
is
a
little
bit
different.
It's
like
I'm
downstream
from
you.
I
get
a
full
handle
on
this
fan
and
I
can
do
all
sorts
of
wacky
stuff,
but
it's
trickier,
obviously
with
htv
clients,
because
it's
like
there
isn't
anything
downstream
from.
C
B
I
mean
like
I
don't
want
to
distract
bigger
and
kelvin.
Actually,
I'm
done.
B
B
So
that's
technically
a
client,
maybe,
but
not
a
not
an
issue
for
us,
because
we
actually
patch
the
appropriate
class
and
not
like
rely
on
notifications.
D
Yeah
notifications
have
gotten
better
in
real
seven
previously
they're
pretty
rough.
I
know
like
didn't,
spend
too
much
time
working
on
trying
to
make
it
a
bit
more
stable
for
previous
versions.
If
you
come
up
with
something
that
actually
seems
to
work
really
well
so
like,
maybe
we
would
have
instrumented
the
orm
more
heavily
with
asn,
but
it
just
was
a
product
of
its
time.
B
I
like
what
we've
done
it's
good
for
this
purpose,
I
so
yeah
I
mean
it
sounds
like
like
I'm
trying
to
think
if
he
could
technically
access
the
that,
like
internal
like
on
the
payload.
There's
that
key
the
open,
telemetry
underscores
under
the
double
underscore
open,
telemetry
underscore
spam
like
key
so
technically,
he
might
be
able
to.
B
Access
that
and
modify
the
spam
name
without
us
necessarily
needing
to
add,
I
don't
know
like
without
us
needing
to
add
the
the
thing,
a
configuration
option
if
we
want
you
know
if
we
want
him
to
use
this
gym,
it
could
also
just
be
like
we
can
take
the
pr
and
say
when
we
add
this
configuration
option,
we'll
update,
because
there's
a
few
that
need
to
get
updated
to
rely
on
this
gem,
and
we
can
just
do
it.
B
You
know
migrate
those
over
when
we
when
we
add
the
config
option,
like
I
don't
know
if
it's
blocking
ariel
left.
So
on
my
I
don't
think,
there's
anyone
who
is
strongly
opposed,
but
yeah,
I
don't
know
anyway.
Those
were
my
things
I
was
thinking
about.
I
think
this
pr
is
really
clo.
I
think
it's
like
probably
should
be
something
that's
done
in
the
next
couple
days.
Chris
has
been
super
responsive.
That's
I've
been
the
blocker
if
anything
per
usual
yeah,
okay.
B
D
B
Checked
race
card
it
doesn't
so
we
shouldn't
make
it
yeah,
but
for
99.9
percent
of
users
who
do
who
will
have
active
support,
it
should
be
fine,
but
anyway
cool
yeah.
I
don't
know
I've
besides,
that's
been
slow
sleepy
on
the
mr
shields.
Has
a
couple
pr's
you
can.
I
haven't
looked
at
it
all.
I
know
other
people
are
looking
at
them.
I
don't
think
there's
been
too
much
going
on
in
control.
D
Yeah
yeah
johnny's
pr
kind
of
put
them
in
front
of
france
last
week,
but
he's
on
vacation,
so
I
gotta
kind
of
pout
him
down
a
bit
because
he
didn't
want
to
look
at
them
specifically,
so
he
has
to
like
hold
on
that.
I
should
have
updated
with
comments
saying
like
holding
for
more
viewers
or
something.
B
I
mean
I've
been
I've.
Looked
at
the
query,
the
graphql
thing,
the
rack
rake
is
pretty
close.
I
think
arielle
had
one
or
two
points
of
feedback.
There
was
one
thing
I
think
it's
a
good
call
out
where
it's
like
there's
an
exit
hook.
They
added
in
this
instrumentation.
B
I
guess,
there's
some
situations
where,
like
it,
I
don't
know
we
you
want
to
ensure
that
some
process
doesn't
get
kept
open
by
or
or
if
it's
like
it.
It
ensures
that
we
would
flush
all
the
spans.
Sorry
we
just
kept
going.
So
I
think
that
was
the
one
book
of
what
all
this
yogurt
the
one
point
of
feedback.
That's
still
there's
still
some
open
discussion
here,
but
this
is
pretty
close.
I
had
yolo
approved
it
per
usual,
but
definitely
still
requires
a
little
bit
of.
B
Yeah
ariel
for
context,
we
just
I
just
went
through
the
other,
open,
pr's
and
kind
of
gave
a
survey
of
where
things
are
at,
which
is
it's
been
relatively
sleepy.
A
couple
things
are
pretty
close.
This
was
one
of
them.
The
contributor
still
has
a
couple
points
of
feedback
they
have
to
address.
I
think.
C
Yeah,
mostly
hoping
to
get
I'm
sure
you
addre
you
address
this
already,
but
getting
maintainer
feedback
around
where
at
exit
hooks
should
actually
be
registered
and,
personally
speaking,
I
think
that
should
be
closer
to
the
sdk
than
it
should
be
to
an
individual
instrumentation.
C
D
I
guess
it
did
well.
I
guess
it
doesn't
have
to
be
hard,
but
it
depends
on
how
you
want
to
have
it
like.
I
don't
know
if
this
is
like
I'm
just
going
off
attention.
You
guys
can
stop
me
if
this
is
useless,
but
we
have
repeated
calls
to
shut
down
throughout
like
internal
deployments
at
shopify,
because
some
of
the
services
that
we
run
like
let's
say
a
job
worker
they'll,
actually
provide
us
a
proper
shutdown
hook
and
so
we'll
use
it.
D
So
it
becomes
a
little
bit
tricky
like.
On
the
one
hand,
it
potentially
doesn't
make
sense
to
have
an
instrumentation,
but
it's
it's
not
totally
clear
to
me
where
it
should
live
in
the
the
sdk,
because
in
the
sdk
I
guess
it
could
be
a
configuration
option.
You
say
at
exit,
but
usually
it's
the
other
direction.
You
want
to
say.
D
I
want
to
shut
down
this
thing
on
this
hook,
so
like
at
exit,
is
like
the
kind
of
like
the
nuclear
option
where
it's
just
like
when
this
rubric
process
kills
we'll
try
to
do
this,
and
I
don't
know
I'm
rambling
now
so
I'll
pause
to
see
if
this
is
actually
not.
C
Because
there's
there's
multiple
layers
to
it
right,
there's
also
term
signals
like
do.
We
want
to
do
something
special
for
term
signals
versus
doing
at
exit,
and
so
I
think
a
lot
of
other
libraries
do
very
sophisticated
things
and
specifically,
like
I'm
thinking
of
the
data
dog,
that's
the
client
that
looks
specifically
for
system
exit
right
and
there's
nothing.
We
can
do
about
exit
bank,
so
I
don't.
I
don't
know
that
there's
any
capabilities
that
we
want
to
provide
from
term
like
by
capturing
term
signals
like
if
we
did.
We
want.
A
B
C
C
C
C
So
when
the
process
is
finished,
the
only
span
we
have
access
to
is
the
parent
rack
span
through
some
sort
of
like
propagation
thing.
So
it's
like
you,
don't
see
where
the
interrupt
actually
happened.
You
don't
see
where
the
actual
last
band
that
was
actually
causing
all
the
latency
was
right
and
that's
that's
not
helpful.
B
Yeah,
I
agree,
I'm
not
I'm
not
saying
that
it's
a
great
state
to
the
world-
I
just
don't
know
one
like.
I
think
tracing
systems
are
by
design
like
lossy.
I
think
there's
some
data
that
gets
dropped
along
the
way
and
like
we're
not
it's
necessarily
like
we're
not
attempting
to
it's
a
best
efforts
approach
and
it's
a
trade-off
between
like
do.
We
want
to
stop
shutdown
for
some
time
while
we
close
and
process
all
these
spans,
which
adds
whatever
you
know
so.
D
Shopify
once
upon
a
time
with
the
previous
tracing
library,
we
would
try
to
flush
out
all
the
spans
when
it
would
shut
down
and
be
like.
D
Let's
get
them
all,
and
we
happen
to
really
break
the
deployment
system
at
shopify,
because
they're
trying
to
deploy
a
new
release-
and
we
said
hold
on
we're
like
still
shutting
down
we're,
still
shutting
down
and
it
got
caught
kind
of
in
a
an
export
loop
because
it
was
creating
new
spans
as
it
was
trying
to
shut
down,
and
so
all
the
processes
would
not
shut
down.
We
couldn't
deploy
new
resources
because
we
had
all
the
computers
trying
to
shut
down.
D
D
It
doesn't
have
the
default,
so
it
was
caught
in
this
export
loop
and
these
job
long
running
jobs
were
stuck
for
like
a
week,
trying
to
shut
down
and
so
and
like
that
was
because
we
used
the
ad
exit
hook
and
it
like
it
did
on
exit
and
it
called
shut
down,
and
it
did
what
it
was
supposed
to
do,
but
because
we
didn't
set
a
timeout
and
we
weren't
using
the
hooks
provided
by
this
job,
rendering
the
drop
runner
framework
we
unintentionally
broke
a
subset
of
their
jaw.
D
Runners
right-
and
I
guess
all
this
like
rambling
is
to
say,
is,
like
I
agree
with
both
eric
and
ariel,
because
I
think,
like
there's
gonna,
be
cases
where
it's
like
you,
don't
wanna
lose
that
information,
but
there's
a
kind
of
like
this.
This
limit
to
best
effort.
But,
more
importantly,
it's
like.
D
D
And
then,
if
you
set
a
timeout,
then
it's
like:
what's
the
appropriate
timeout,
it's
like,
then
it
becomes
configurable
and
it's
like
what,
if
this
shuts
down
earlier
than
something
else,
and
you
would
like
to
have
an
extra
time.
So
it's
like
everybody's
now
becoming
responsible
for
shutting
this
thing
down,
or
do
we
say
it's
only
for
rack
or
maybe
it
should
be
instrumentation
for
puma.
D
Like
I
don't
know,
it's
very
very,
I
think,
subjective
of
where
these
things
should
live.
So
I'm
I'm
weary
of
including
it
in
anything.
Sorry,
cop
out
answer.
A
Yeah
my
my
two
cents
on
this-
I
guess,
is
that
for
for
something
like
rake,
you
do
need
to
be
able
in
like
these
short-lived
processes.
You
do
need
a
way
to
be
able
to
flush
out
these
fans
at
the
end.
Otherwise
you
know
the
instrumentation
is
not
really
worth
having.
So
I
am
kind
of
of
of
the
opinion
that
this
should
be
an
sdk
concern,
though
there
should
be
like
some
sort
of
graceful,
shutdown
down
and
flush.
A
You
know
some
sort
of
mechanism
that
is
there,
but
based
on
these
discussions
here,
it
seems
like
it's
not
like
a
one
size
fits
all,
so
there
should
be
some
knobs
on
it
to
like,
say
you
know
what
actually
I
just
want
you
to
shut
down
as
fast
as
possible
and
I'll
accept
losing
that
last
batch
of
spans,
but
I
think
we
should
probably
have
this
centralized
at
the
sdk
level
with
appropriate
configurations
to
tune
it
to
your
specific
needs.
A
B
Yeah
yeah
back.
So
what
should
we
say
to
the
pr
guy.
A
So
in
the
end,
would
that
end
up
being
like
some
suitable
shutdown
method
that
we
that
we
like
for
the
sdk
and
would
that
would
the
sdk
registered
at
exit
hook?
That
calls
that
more
or
less
is
this?
What.
B
B
Yeah,
I
wonder
if
we
can
give
him
a
middle.
I
don't
like
doing
these
things
where
it's
like.
Well
we're
going
to
do
this
eventually.
So,
let's,
let's
merge
in
something
we
don't
like
in
the
meantime,
because
we
don't
want
that
to
block
your
thing,
but
I
think
at
the
very
least,
if
we
do
take
it
out
of
exit
year,
it
would
need
some
sort
of
timeout
or
some
more
robustness
would
be
just.
C
B
C
B
They
they've
added
a
couple
instrumentations,
if
I
recall
correctly,
so
I
think
they're
users
currently
so
cool.
I
think
that's
a
good
middle
ground
solution
for
now
is
like
look
just
manage
this
yourself.
Here's
a
we
can
provide
a
snippet,
be
like
here's
one
example
of
how
to
do
this
and
you
know
whatever
we
can
do
them
a
courtesy
of
picking
them.
If
we,
if
we
add
a
an
sdk
convenience
helper
for
it
in
the
future,
feels
like
everybody
wins
free
lunch
for
everybody.
A
As
part
of
all
this,
just
kind
of
make
make
an
issue
for
for
an
sdk,
graceful,
shutdown
or
comparable.
That
would
be
the
the
ultimate
nsaid
and
then
yeah.
B
A
B
Yeah
it'll
also
be
a
good
central
place
for
us.
Like
I
have
you
know
my
thoughts
on
this
are
literally
what
you've
heard
out
loud
in
the
past.
Three
minutes
like
I
don't
have
deep
thought
like.
So,
if
we
it's
a
good
central
discussion,
if
we
later
realize
that's
a
terrible
idea,
it
would
be
good
to
which
maybe
it
is
anyway
cool.
I
can
add
an
issue
on
the
main
repo
and
I
will
link
to
it
in
this
I'll
I'll
run
with
the
action
items
from
this.
C
I
know
I
just
I
just
I
just
converted
a
comment
to
a
nation,
I'm
about
to
transfer
it
to
to
the
upstream
and
you
can
add
more
color
to
it
if
you'd
like,
but
it's
like.
D
That's
a
good
segue.
Have
you
done
any
tracing
and
github
actions
like
have
you
traced?
It
have
actions
themselves
yeah.
D
Like
if
I
was
a
user
of
github
actions,
would
you
be
able
to
provide
me
guidance
of
tracing
these
github
actions
because.
D
C
Vendors
provide
actions
for
tracing
integration,
but
github
does
not
provide
the
ability
for
you
to
hook
into
that
like
into
our
internal
tracing.
C
B
Alex
but
I
don't
know
his
real
name,
I
have
action
to
otlp,
I
don't
know
what
it
does.
I
just
saw
some
tweet
and
it
was
like
cool
and
that
was
like
two
months
ago
and
in
my
mind,
just
code
boat
and
stuck
so
I
don't
know
what
the
I
I
assume
it's
like
some
like
that
you
know
unopinionated
dag
at
exports,
and
you
can
turn
it
into
a
trace.
If
you
want
or
something
I
don't
know
all
right.
D
B
D
B
B
A
A
Take
a
look
at
this.
If,
if
you
have
any
questions,
I
do
have
access
to
code
bulletin
feel
free
to
pass
things
to
me.
It
looks
like
it
looks
like
it's
nearing
a
year
old,
so
we'll
see
if
it
is
it's
held
up
over
time,
but
I
know
it
was
something
he
was
experimenting
with
previously.
D
D
I
can
quickly
summarize
it
so
that
we
don't
have
to
dig
through
it
once
upon
a
time
there's
graphql
instrumentation.
There
are
some
fields
that
are
more
verbose
than
others
like
platform,
key
field
resolution
and
something
like
optimize
something
field,
those
all
those
fields
create
incredibly
long.
D
Traces
for
graphql,
like
I
do
mean
like
it,
goes
from
like
100
to
like
11
000
in
one
case
that
I
was
testing
on
a
production
app,
so
I
added
some
config
options
to
either
turn
on
or
turn
off
these
things
they
default
off,
but
to
the
developers
who
work
on
graphql
and
write
resolvers
and
all
those
fun
things
like
there's
there's
value
to
be
had
in
those
fields,
but
I
can
confidently
make
a
case
for
my
workplace
that,
like
on
all
the
time,
is
not
a
reasonable
option.
D
So
I
think
we
can
get
to
the
point
of
there
is
some
reasonable
desire
to
have
this,
like
maybe
per
request,
like
I'm
digging
into
something
I'd
like
to
turn
on
these
noisy
fields.
For
this
one
request,
so
I
can
look
into
my
problem
great,
so
tony
matthews
had
submitted
a
pr
to
actually
the
actual
graphql
gem
to
add
support
in
some
ways,
there's
a
bit
of
a
disconnect
between
how
config
works
and
it
didn't
work
as
intended,
or
it
wouldn't
work
as
intended.
D
D
D
Does
this
approach
support
that
if
you
scroll
down
there's
a
comment
by
rob
who,
if
I'm
reading
it
correctly,
seems
to
indicate
that
that
is
possible?
This
approach,
I
need
to
actually
dig
into
that,
because
they're
not
familiar
enough
with
craftql
yeah
ariel's.
Getting
to
the
point
is
that
I
think
this
problem
should
be
solved
with
baggage.
So
I
think
that
we
should
add
kind
of
an
e-noun
configuration
for
the
instrumentation.
D
It
would
look
like
on
off
or
per
request
when
it's
on
it's
on
when
it's
off
it's
always
off
per
request,
means
that
it
will
check
for
the
presence
of
some
baggage
key.
What
the
key
would
look
like
again
is
going
to
be
up
to
a
lot
of
subjective
debate,
because
it's
inspect,
I
think
in
the
more
complex
version
of
this
you
could
specify
which
service
and
which
key
as
part
of
the
baggage
and
you
could
specify
like
every
second
service.
D
So
again
rob
if
you
scroll
down
further
past
me.
He
he's
another
employee
at
shopify
and
he
works
a
lot
on
like
graphql,
ips
and
checkpoint
he's
very,
very
knowledgeable
in
space.
He
makes
a
good
point
that
graph
fuel
is
not
necessarily
invoked
through
an
issue
request
which
is
actually
news
to
me
by
the
way.
I
didn't
know
that
so
that's
a
pretty
valid
point
of
saying,
hey
like
we
might
not
be
able
to
do
this,
but
I
still
think
baggage
is
the
way
to
go
here.
C
Yeah
there
there's
a
there's
a
lot
of
like
nebulous
areas
here
right
so
like
we.
If
we
start
introducing
the
notion
of
some
sort
of
client-side
verbosity.
D
I
know
that
this
will
deliver
a
lot
of
value
to
the
people
who
care
about
the
graphql
apis
that
they
maintain,
I'm
afraid
that
this
is
going
to
get
caught
in
spec
limbo,
and
it's
like,
I
almost
have
a
small
inclination
to
be
like.
Do
we
fork
this
instrumentation
internally
and
just
do
the
thing
we
need
to
do
to
make
it
work.
D
D
C
Without
building
consensus,
though,
where
are
we
going
to
be
right
so
like,
if
you're?
If,
if
we're
starting
this
implementation,
then
we're
going
to
have
to
have
other
languages
on
the
client
side?
Do
something
about
it
in
order
for
us
to
get
like,
you
know
lucky
for
us
right
that
we
control
our
ruby
applications,
but
if
somebody
wants
to
use
this
capability,
they're
not
going
to
see
it
in
other
implementations,
you
know
what
I'm
saying
you
know
this
goes
back
again.
C
C
This
is
what
we're
doing
at
shopify,
and
this
is
the
reasons
why
we're
doing
it,
because
and
because
you
know
that-
is
this-
something
that,
like
the
the
future
control
plane,
should
handle
really
because
there's
a
lot
of
things
like
that,
I'm
interested
into
you
know
you
go
back
to
like.
I
want
to
be
able
to
say.
I
want
to
ensure
that
we're
capturing
the
trace
for
this
specific
client
through
a
feature
flag
and
give
me
as
much
verbosity
in
all
of
these
instrumentations,
but
for
99
of
my
use
cases.
C
I
don't
care
to
capture
the
traces
for
some
of
these,
or
so
I
don't
care
to
capture
these
fans,
but
I
really
want
to
crank
it
up
for
this
one
user.
I
want
to
see
everything
they're
doing
and
the
only
way
to
do
that
now
is
through
some
sort
of
like
tail
sampling
through
the
attribute
processor
right
and
doing
that
through
the
collector.
C
D
C
C
B
I,
like
the
approach
jen,
you
know
just
like
speaking
without,
like
outside
of
context
of
knowing
the
the
work
required
to
get
some
and
the
actual
obstacles
around
spectrums.
Like
I
think
it's
the
right
approach,
I
think
using
baggage
is
appropriate.
It's
a
nice,
it's
a
creative
use
and
like
it
works
and
yeah.
I
don't.
C
B
D
D
D
Diversions,
we
don't
need
yeah,
we
are
we're
over
now,
so
I
think
it's
tricky
like
here.
I
think
this
does
require
follow-up
conversations
and.
D
Prototype
something
that
doesn't
require
spec
definition
that
can
be
used
like
can
we
if
we
make
use
of
all
the
tools
we
have
available
to
us
via
samplers
and
this,
and
that
can
we
do
this
without
having
to
formally
define
it
and
without
having
to
make
it
a
configuration
option?
Like
is
this?
Can
we
get
to
that
point?
So
I
need
to
spend
a
little
bit
more
time
with
this,
but
I
also
wanted
to
put
in
front
of
other
people,
because
I
think
it's
useful.
D
I
think
the
idea,
the
intent
of
this
pr
is
very,
very
practical
and
very
useful
to
people
who
are
actually
developing
systems
with
graphql
apis
like
this
is
coming
from
our
kind
of
like
internal
team
that
maintains
a
lot
of
the
graphql
apis
at
shopify.
So
it's
like,
I
think,
it's
a
good
like
a
very
battle
tested,
kind
of
perspective.
That's
like
asking
for
this,
so
I
think
that's
interesting.
I
think
there's
value
that
not
just
being
like.
Oh
shopify,
does
it
so
it's
useful.
D
B
Yeah
we're
end
users
yeah
yeah,
I
mean
it'd,
be
cool
to
see.
If
it
could
get
done
with
the
tools
available,
you
can
touch
baggage
and
a
sampler.
B
You
know,
I'm
sure
it
would
be
a
little
look
a
little
weird,
but
this
is
a
good
discussion.
I
appreciate
you
bringing
it
up.