►
From YouTube: 2022-05-24 meeting
Description
cncf-opentelemetry meeting-2's Personal Meeting Room
A
C
Yeah,
it
looks
like
yeah
congratulations
on
the
award.
D
I
get
it
every
year
on
father's
day,
so
it's
really!
D
D
D
It
is,
it
is
heavy
as
ahead
of
where
the
crown
hopefully
this
year,
you
know
I
can
pull
it
out
again,
but
it's
a
lot
of
stiff
competition
out
there
with
a
lot
of
dads
working
from
home.
D
D
Oh,
that's:
that's
going
that's
going
to
go
on
youtube,
hopefully
that
I.
E
But
no,
I
am
I'm
struggling
with
internal
terraform
kubernetes
management
things.
I
have
grand
ambitions
and
projects
but
have
been
stymied
by.
E
How
do
you
actually
update
something
in
a
large
environment
ever
as
it
were,
but
I'm
okay,
otherwise,
I'm
visiting
with
a
friend's
cats
they're
out
of
town
and
I
was
feeding
the
cats
and
the
cats
were
you'd-
think
they'd
been
left
alone
for
weeks
on
end,
the
way
that
they
were
attacking
me,
but
no,
it's
been
less
than
24
hours,
they're,
just
very
needy
one
of
them
bit
me
after
I
stopped
petting
it,
and
I
I
I'm
trying
to
figure
out.
E
D
Well,
I
don't
know
when
you
walked
in
the
door,
they
greeted
you
with
shurikens
and
nunchucks.
So
I
imagine.
F
F
Yeah
so
I
mean
I
think
I
attended
one.
Maybe
two
years
ago
the
group
was
slightly
different
back
then.
I
think
there
were
about
two
people
and
then
kind
of
switched
projects
at
my
company
for
a
while,
but
back
on
observability
engineering,
as
it
were
so
deep
into
it
all
and
quite
interested
in
joining
in.
D
Okay,
well
welcome
to
our.
I
don't
know,
you're
a
veteran
right,
so
you
should
be
welcoming
all
of
us
newbies
who
have
been
around
for
two
years,
but
I'm
marielle
and
I
work
at
github
and
I'm
on
the
observability
team.
Here.
D
My
my
colleagues
down
there
are
I'll
pop
corn
over
to
a
sam
handler
up
next,
so
he
can
introduce
himself.
A
C
Cool
yeah
I
might
have
been
around
when
you
came
last
time,
but
in
any
case
I'm
matt.
I
work
at
lightstop
welcome
back,
headed
off
to.
G
Robert
matt
is
like
the
og
of
og's,
I
think
for
open,
puncture,
ruby
him
and
francis
who's,
another
shopify
person
who
isn't
on
the
call
I'm
also
at
shopify.
I've
been
working
on
this
for
a
little
while
now,
yeah,
welcome
glad
to
see
anything
or
an
og
new
phase.
F
Yeah,
it
wouldn't
really
retain
the
title
of
og
because
I
didn't
actually
contribute.
It
was
more
just
an
interest
at
that
time,
but
I
guess
I'll
go.
I'm
nick.
I
work
at
a
company
called
zappy.
We
do
market
research
which
might
not
be
particularly
interesting
to
anybody
and
yeah
senior
software
engineer.
We
are
ruby
on
rails
house,
for
the
most
part
have
a
bit
of
python
go,
but
that's
pretty
much
where
my
expertise
is
cool.
Welcome.
E
I
was
just
going
to
see
if
people
were
going
to,
let
me
get
away
with
it
too.
My
name
is
andrew
hayworth.
I
also
was
not
around
when
you
were
joining
the
meetings
before
currently.
I
am
at
shopify
doing
observability.
Previously
I
was
at
github
with
audiel
doing
observability,
and
then
there
was
a
brief
interregnum,
where
I
was
at
a
startup
that
we
don't
really
talk
about
now,
where
I
was
also
ostensibly
doing
observability.
C
Cool
yeah,
so
we
can
go
ahead
and
and
get
out
of
the
meeting.
Typically,
the
format
is
I'll.
Try
to
quickly
recap:
the
sig
been
trying
to
keep
it
to
like
10
minutes
or
so
usually
fail
on
that
point,
but
trying
and
then
kind
of
move
it
over
to
questions
a
little
bit
more
relevant
to
our
repository.
C
C
C
There
is
basically
a
discussion
happening
on
how
to-
or
I
guess
it's
a
spec
pr
on
guidance
for
the
additive
property
of
metrics,
and
basically
it's
just
kind
of
going
through
a
number
of
examples
where,
if
you
have
a
cumulative
metric,
if
a
back
end
kind
of
receives
a
value,
it's
receiving
a
stream
of
values,
and
then
one
of
them
comes
in
less
than
a
previous
value.
It
kind
of
indicates
that
a
reset
has
occurred
for
certain
instruments
for
like
a
yeah
for
for
additive
measurements.
That's
true
for
non-additive
measurements.
C
C
C
To
these
repeated
key
values
to
instrumentation
scope,
so
apparently
there
will
be
some
spec
spec
updates,
incoming
which
I
guess
will
translate
into
some
updates
for
hotel
ruby.
I
think
you
know
the
big
thing
that
we
were
talking
about.
We
were
talking
about.
This
was
like
how
does
this
affect
like
tracer
initialization,
because
he
seemed
to
be
bound
to
a
tracer
at
get
tracer
time.
I
think.
E
Oh
yeah
that
got
merged
they
asked
about.
If
whether
or
not
we
could
implement
that
whenever
we
get
around
to
actually
switching
to
instrumentation
scope,
and
the
answer
was
yeah
sure.
Why
not
that's
another
hash,
since
this
is
ruby
where
memory
is
cheap
and
we
pretend
that
everything
is
even
though
it
usually
is
sorry,
I'm
in
a
mood
today
but
yeah.
This
seems
like
it's
pretty
straightforward,
just
bind
it
to
the
scope
when
you
get
the
tracer
and
it
goes.
C
C
We
may
want
to
add
configuration,
I
guess,
on
a
per
instrumentation
basis
to
add
scope,
attributes.
But
that's,
I
guess,
we'll
scope
this
out
as
the.
C
There
is
some
tooling
to
help
generate
a
metric
stable
from
from
yaml.
This
is
there's
already
some
tooling
for
the
semantic
conventions
just
for
like
trace
and
resource
attributes,
but
metrics
have
been
a
little
more
complex.
So
if
you've
been
waiting
for
something
like
this,
it's
here.
C
C
C
Make
sense
of
a
lot
of
it
and
ingest
it
successfully,
but
there
may
be
some
data
that
it
doesn't
know
to
do
with
so
giving
the
backend
a
way
to
say
that
it
has
partially
ingested
the
data
rather
than
just
like
a
success
or
failure
is
kind
of
the
goal
here.
I
think
right
now
the
proposal
was
just
to
kind
of
like
return,
an
error
message,
and
I
know
discussion.
C
Josh
was
kind
of
commenting
in
person,
so
I'm
not
sure
what
exactly
he
wrote
here,
but
he
wanted.
He
wanted
more
data.
He
was
just
saying
that
nobody
knows
what
to
do
with
a
string
error
message
other
than
login
and
well,
maybe
slightly
better
than
then
just
like
success
or
failure.
C
He
was
hoping
to
have
something
a
little
bit
more
like
like
structured
log,
that
you
could
extract
some
more
useful
data
out
of,
I
think,
I'm
not
sure
exactly
what
form
that
would
take,
but
it
looks
like
there's
various
proposals
to
try
to,
like
you
know,
get
as
much
information
back
to
the
sender
about
what
was
okay
and
what
was
not.
C
And
that
was
that
was
a
recap.
Any
questions
comments
concerns.
D
D
D
C
Yeah,
I
think,
we'll
see
where
that
discussion
goes.
I
think,
probably
most
likely.
The
the
goal
of
this
is
to
be
able
to
indicate
to
a
client
like
what
things
you
might
if
you
sent
again
might
have
a
chance
of
getting
ingested
and
things
that,
if
you
send
it
again,
are
just
gonna
continue
to
fail.
So
you
can
a
retry
strategy
of
some
of
some
kind.
You
can
optimize
it
as
much
as
possible.
C
It
seems
like
there
are
no
burning
questions
listed.
Are
there
any
reporting
questions
prs
issues.
D
So,
to
kind
of
like
to
kind
of,
like
summarize,
we
kind
of
like
you
know,
we
introduced
this
idea
of
allowing
people
to
configure
their
like
configured
trace
propagation
on
the
consumer
side
right.
So
you
can
tell
the
consumer
to
use
a
child
to
continue
the
traces
if
it
were
a
child
span
to
use
a
link
or
to
ignore
it
all
together,
and
in
retrospect
you
know,
like
I
mean
I
brought
up
some
concerns.
D
D
So
I
don't
know
if
anybody
has
any
familiarity
with
the
messaging
spec
portion
or
what
changes
are
going
through
there.
I
really
would
appreciate
some
participation
in
the
conversation
in
here,
because
eric
was
the
one
that,
like.
D
You
know
spent
some
time
chit-chatting
through
this,
but
not
participating
much
anymore,
so
we
kind
of
lost
like
our
area
of
like
expertise
in
that.
One
thing
I
will
propose,
though,
is
for
us
to
get
rid
of
the
none
option,
because
I
think
that
will
confuse
people
I'm
trying
to.
I
was
trying
to
think
of
a
use
case
of
where
we
didn't
want
to
connect
a
trace
either
through
a
link
or
a
child,
and
I
couldn't
think
of
one
the
only
other.
D
The
only
case
that
I
could
think
of
was
you
know
we
had
introduced
it
in
the
case
where
maybe
you
have
like
a
big
batch
job
that
kicks
off
a
bunch
of
little
jobs
and
and
batching
semantics
aren't
really
in
the
messaging
specification,
like
that's
still
kind
of
ambiguous.
E
I
think
you're
right.
If
I
remember
correctly,
you
and
I
talked
a
lot-
a
lot
about
this
internally
at
github
and
we
were
instrumenting
some
of
the
stuff,
and
that
is
one
of
the
cases
where,
like
some
action
in
the
github,
ui
kicks
off
a
million
batch
jobs
and
it's
actually
just
not
useful
to
have
propagation
there,
and
I
think
we
might
have
eventually
added
link
because
we
wanted
some
sort
of
reference.
But
the
idea
that
you
just
don't
want
them
connected
because
of
what
your
system
does.
I
think
you're
right.
D
D
D
No,
no
I'm
looking
for
some
clarification
from
amir,
because
I've
read
some
things
and
I
thought
I
understood
you
know
you
meet
some
cases,
the
case
being
that
look,
we're
doing
something
that's
not
defined
by
the
spec,
and
I
get
that.
A
D
And
so
he
didn't
want
to
do
something
that
was
not
defined
by
the
spec
and
describe
some
instances
of
which
that
would
be
a
problem,
and
I
think
none
if
we
eliminated
none.
That
would
be
one
of
the
problems,
but
I
think
also
to
kind
of
like
satisfy
our
user
base,
which
most
folks
are
like.
I
really
want
to
see
my
background
jobs
connected
to
my
web
requests
and
our
backend
systems.
Don't
really
support
it.
D
I
think
that's
the
reason
why
that's
a
driving
motivator
for
supporting
child
right,
because
this
was
our
experience
with
open
tracing
where
these
vendors
didn't
implement,
follows
from
very
well
if
they
supported
it
at
all.
Right
so
like
that
was
my
case
for
wanting
to
at
least
keep
child,
but
I
also
don't
use
aw,
you
know
sqs,
so
it
may
not
impact
me,
so
I
may
not
give
a
we're
just
doing
it
for
like
symmetry,
for
because
other
instrumentations
did
the
same
thing.
E
There's
also
if
we
can
get
the
clarification
use,
cases
for
for
this,
behavior
is
going
back
to
the
spec
and
talking
about
and
identifying
these
different
cases,
I
feel
like
people
use
message
so
message:
queue
systems
for
different
purposes
and
the
shape
of
your
instrument.
The
shape
of
your
traces
coming
out
depend
on
what
you're
using
what.
E
Why
are
you
enjoying
a
message
and
what
picks
it
up
and
as
nicholas
points
out,
if
you
have
a
a
million
things
in
a
batch,
you
probably
don't
want
to
spam
for
every
one
of
them,
or
maybe
you
do,
but
you
want
to
like.
E
You-
want
informed
consent
on
that
of
your
spans
if
your
traces
are
going
to
turn
into
into
a
waterfall.
But
all
of
this
to
say,
if
we.
A
E
D
Right
so
so,
in
the
cases
where
we
we,
where
we
were
introducing
the
idea
of
propagation,
none,
meaning
that
you
were
going
to
do
your
own
manual
propagation
yourselves
and
opt
into
specific
propagations
yeah,
because
there's
some
cases
where
it's
like.
I
do
care
that
this
message
that
I'm
in
queueing
gets
picked
up
and
there's
only
certain
things
that
I
want.
E
Right
versus
more
like
fire
and
forget
stuff,
where
maybe
you
sometimes
you
want
to
continue
the
trace?
Sometimes
you
want
to
make
a
new
trace
but
link
them
and.
F
D
E
So
but
I
think
that
was
a
lot
of
words
to
say.
I
think
I
agree
with
you
that
we
could
use
more
information
about
the
use
case
so
that
we
we
maybe
keep
our
offscript
implementation
and
and
maybe
go
back
to
the
spec
and
argue
for
you
have
options
when
doing
message
cube
instrumentation,
so
that
the
spec
says
that
you
have
options.
And
then
maybe
we
have
conversations
about
when
you
use
which
option.
D
E
Yeah
well,
which
I
missed
it,
which
sig
was
eric
attempting
that
would
have
more
context
on
this
problem.
D
Wasn't
it
attending?
He
just
read
all
the
documentation
because
he
worked
on
the
messaging
stuff
for
for
datadog
when
he
was
there,
so
the
I
don't
think
it's
the
instrumentation
stick.
I
think
there's
a
special
messaging
sig.
D
C
There
is
yeah,
there's,
there's,
definitely
a
message
in
sig.
Apparently
they
meet
every
thursday
at.
C
C
I
feel
like
I
feel,
like
amir,
has
gone
to
a
lot
of
those
who
is
one
of
yaniv's.
Oh
okay,
here
we
go
co-workers
but.
C
It
would
be
good
to
know
like
what
the
current
state
of
things
that
are
currently
specified
are
like
are
they
are
there
yeah
are,
is,
is
any
of
the
current
spec
being
heavily
refactored?
I
think
is
one
thing
that
would
be
useful
to
know
or
should,
or
can
we
kind
of
rely
on
what
is
there
as
being
pretty
stable
and
the
stuff
that
they're
doing
is
more
like
some
bite
shedding.
F
It's
funny
you
should
ask,
I
wouldn't
say
necessarily,
but
I
do
need
to
add
something
to
the
aws
sdk
instrumentation,
because
right
now
it
doesn't
actually
support
the
architecture
that
we
have.
We
have
a
polar
which
fires
at
a
queue,
so
creating
infinite
amounts
of
spans.
Even
if
there
aren't
messages
present
on
the
queues.
F
Just
really
a
waste
of
resources
for
us
do
have
a
little
bit
of
expertise
on
the
messaging
side,
though
we
do
use
it
a
little
bit.
I
would
not
say
that
we're
full-blown
message
queues,
but
there
are
parts
of
our
application
where
we
do
use
it
as
opposed
to
http.
For
consistency's
sake,
I
could
start
attending.
I
don't
exactly
know
what
what
my
objective
is
so
to
speak.
I
still
need
to
get
quite
comfortable
with
the
sixth
as
it
is,
but
it
does
interest
me.
A
Yeah,
I
was
just
gonna
say
I'm
happy
to
go
this
week
and
try
to
rattle
some
cages
and
be
like
hey.
Can
you
look
at
this?
I
don't
if
anyone
else
like
nicholas,
if
you're,
if
you're
desperate
to
get
into
that
sig
by
all
means,
take
it,
but
in
the
interest
of
pushing
this
stuff
forward,
I'm
happy
to
go
this
week
and
be
like
hey.
Can
you
look
at
this
pr
and
tell
us
what's
going
on?
I
don't
mind
at
all,
so
unless
someone
tells.
A
D
You
don't
necessarily
have
to
attend
the
sig,
but
rather
like
understand
the
outcomes
of
what
happened
and
spend
some
time,
maybe
because
the
cigs
are
recorded
also,
so
if
you
can't
attend
in
person
or
whatever
to
watch
them,
that's
all
I
that's
what
I'm
hoping
for
so
that
at
least
we
can
present
something
to
them
and
say
look.
This
is
what
we're
doing,
and
this
is
where
we
see
the
gaps,
and
this
is
what
our
our
users
request
from
us.
D
So
other
people
experiencing
the
same
thing
like
I'm
sure
somebody
out
there
is
having
a
very
similar
experience.
Otherwise,
like
okay,
I
don't
think
it'll
be
unique
to
ruby
on
rails
users.
You
know
to
say
I'd
like
to
see
my
active
record
jobs
connected
to
the
http
request
that
came
in.
I
don't
think
that's
unusual.
D
So
that's
that's
that's
more
about
what
my
expectations
are.
I
guess.
A
Or
sam
you're,
gonna
say
something
no,
I
was
gonna
say
it
sounds
like
nicholas
might
be
interested
in
that
sort
of
I
was.
I
was
volunteering
as
a
one-time
when
one
time
ask
a
question
kind
of
guy,
but
if
some
nicholas
you
want
to
read
the
spec,
it's
all
yours.
F
Yeah
yeah,
I
mean,
if
you
absolutely
just
test
message
queues,
then
I'm
happy
to
take
the
take
the
reins
on
that
one.
I
think
it'll
be
good
for
you
to
maybe
go
fight.
The
good
fights
with
this
pr
here
and
kind
of
just
bring
the
conversation
forward,
but
over
the
long
term,
I'd
be
happy
to
kind
of
take
the
reins
there.
I
can
also
just
join
this
week
as
I
get
settled
in.
C
Yeah,
so
it
sounds
like
it
sounds
like
we
kind
of
have
two
things
to
consider
here.
One
is
to
get
some
advice
from
that
group
about
this
particular
issue
and
kind
of
about
this
feature
that
we
kind
of
added
a
while
ago.
I
guess
where
we
kind
of
are
allowing
users
to
select
whether
it's
going
to
be
link
child
or
none,
because
it
seems
like
I
don't
know.
We're
probably
gonna
have
to
do
something
about
that
at
some
point.
C
We're
gonna
have
to
come
up
with
a
plan,
because
I
think
this
is
going
to
create
a
little
bit
of
friction
with
people
working
in
other
languages
that
may
or
may
not
have
these
options.
C
I
went
to
that
sig
a
long
long
time
ago,
when
it
first
kind
of
spun
up
just
to
kind
of
get
an
idea
of
what
was
going
on
there
and
they
were
bringing
in
there
were
some
people
who
were
like
kind
of
like
you
know,
experts
on
on
message,
key
and
message:
q
internals,
so
I
feel
like
they
were
really
thinking
about.
C
Like
you
know,
things
like
kafka
and
trying
to
understand
like
what
exactly
is
going
on,
you
know
under
the
hood,
when
you
try
to
enqueue
or
receive
a
message,
and
it
was
a
little
bit-
I
don't
know
I
feel
like
they
were
zoomed
in
a
little
bit
more
then
I
want
to
connect
my
my
active
job
jobs
to
a
trace,
so
I
I'm
not
sure
how
much
that
higher
level
use
case
has
been
advocated
for
or
even
talked
about
in
that
group.
C
So
but
yeah
it
sounds
like
so.
I
guess
yeah
on
one
hand,
if
sam,
if
you
wanted
to
go
to
try
to
get
some
advice
on
this
particular
pr.
I
think
that
would
be
useful
and
then
sounds
like
nicholas
you
have.
You
have
use
cases
and
things
that
you
would
like
to
see.
So
you
don't
end
up
with
1
million
span
traces,
more
or
less,
and
I
think
having
that
as
a
having
your
feedback
here,
I
think
is,
is
quite
timely.
C
Just
to
make
sure
that
that
information
at
least
gets
surfaced
to
this
sig,
and
also
that
with
this
pr,
we
can
accommodate
that
in
some
reasonable
way.
F
C
C
C
There
are
a
lot
of
links
going
on
in
the
chat
mainly
to
the
actual
meeting.
C
D
Nothing
any
any
other
pr's.
I
think
folks
are
saying.
G
Yeah
so
I
kind
of
parked
merging
some
of
the
approved
things
because,
like
we
started
that
release
fiasco
last
week
and
merging
merging.
G
D
G
Open
releases
are
out
there,
I
find
it
just
makes
the
entire
process
a
bit
more
cumbersome.
So
I
was
like
okay,
let's
get
some
releases
out
and
then
we'll
get
the
merch
train
going
again,
but
the
release
train
never
started
we're
still
at
the
station.
I.
D
Robert
that
I'm
sorry
matthew
that
1275
there's
a
big
old
red
x
on
that
one.
Would
you
mind
clicking
that
big
old
red
x.
G
I
don't
know
so
after
this
release.
We
get
this
one
sorted
I'll
get
some
of
the
merging
going
again
and
then
we
can
start
the
release.
Train
does
again
sound
okay,
ariel
because
it's
kind
of
I've
been
just
trying
to
unblock
you
as
much
as
I
can
with
all
that
stuff
and
keep
things
flowing,
because
I
know
you're
doing
a
lot
of
work
with
the
split.
So
is.
D
G
D
D
I
kind
of
wish
we
stayed
with
single
release
version
numbers
and
did
just
continue
doing
big
bang
releases
for
all
of
them
for
a
little
while,
because
it
was
like
at
least
for
the
instrumentations
and
because
it's
becoming
a
little
unmanageable,
unmanageable
to
like
one-off,
one-off
updates,
all
the
time
we
like
miss
all
the
time.
It's
driving
me
nuts,
but
that's
for
another
day
for
now
I'll.
Try
to
get
these
bills
working.
D
And
I'm
going
on
vacation
next
week,
so
if
I
don't
get
it
done
by
thursday,
there's
nothing
getting
done
next
week.
I'm
sorry.
G
That's
right,
don't
need
to
be
sorry,
I'm
hoping
next
week.
I
can
do
another
update
on
the
metrics
work.
I've
been
chugging
away
at
that
as
much
as
I
can
with
some
other
things
that
have
been
pulling
my
attention,
but
right
now
I'm
really
dialed
in
on
kind
of
the
internal
state
model.
G
I
feel
like
once
that
comes
together
I'll,
be
at
a
better
place
where
I
can
be
like
okay
merge
what
we
have
I'll
need
to
update
some
changes
based
the
spec
change
since
I
started
this
pr,
so
there's
some
things
that
need
to
happen
to
kind
of
get
things
more
in
line,
and
then
I
think,
from
there
it'll
be
kind
of
we'll
be
in
a
good
position
that
like
if
anyone
is
feeling
a
little
masochistic,
they
can
like
jump
in
and
join
me
and
kind
of
work
on
different
pieces
of
this.
G
But
I
think
before
this,
like
core
internal
state,
is
solidified
having
multiple
people
trying
to
collaborate
is
just
going
to
be
miserable.
So
I
kind
of
like
to
comment
on
the
pr
it's
not
necessarily
worth
reading.
Basically,
it
says
like
here
are
the
bits
I
want
to
work
on
and
then,
after
that
we
can
probably
get
some
help
if
people
want
to
say
like
we're
going
to
export.
If
someone
wants
to
work
on
a
prometheus
exporter,
I'm
going
to
try
to
stick
my
claim
to
the
otlp
metrics
exporter.
G
I've
started
plumbing
it
through
locally.
I
actually
have
this
like
really
janky
local
flow,
where
I
can
emit
metrics
and
it
hits
a
collector
which
is
pretty
cool,
so
that's
kind
of
an
update.
I
know
people
have
asked
and
I
don't
know
if
you
anyone
here
has
been
wondering
themselves,
but
it
is
like
actively
being
developed,
even
if
it
looks
kind
of
sleepy
on
the
pr.
D
G
D
All
these,
these
prometheus
style
metrics,
and
these
like
batch
down
metric
the
aggregates
and
exemplars
like
what
do
I
have
a
vendor
that
even
supports
supports
that
log.
D
G
Yeah
we're
using
stats
the
instrument
which
maybe
give
it
away
from
the
fact
that
that's
like
a
shopify
repo,
but
I'm
embarking
on
the
arduous
task
with
a
few
other
people,
including
andrew
to
potentially
not
use.
That's
the
instrument.
D
Yeah
yeah
no
clue
anyway,
yeah
I'm
in
the
no
clue
group
about
even
with
prometheus
style
metrics,
I'm
like
what.
D
I
was
told
there
would
be
no
math,
that's
great,
oh
rob
digging
at
me.
Old
soldiers
is
coming
out
of
github
actions.
One
of
these
days.
Well,
I'm
sure
that'll
be
an
offering.
D
C
Cool,
yes,
thanks
for
the
update,
it's
good
to
see
that
there's
some
progress
being
made
there
and
like.
I
think,
if,
if
I
remember
kind
of
where,
where
we
were
the
last
time
we
updated
this,
like
you
were
really
there's
like
so
many
pieces,
you
need
to
kind
of
get
in
place
to
get
like
the
whole
pipeline
of.
C
But
until
that
point
it's
like
there's
just
so
much
so
much
of
the
pipeline
that
needs
to
be
built.
It
gets
hard
to
have
multiple
people
working
on
it.
Is
that
an
accurate
summary
yeah.
G
Yeah,
especially
yeah,
basically
I
want
to
make
it
so
there's
like
plug
points
like
even
the
asynchronous
instruments.
I
think
that's
something
that
can
be
like
carved
off
once
this
part's
done,
because
it's
like
okay,
we
know
what
synchronous
instruments
look
like.
So
how
do
we
do
that
and
yeah
like
I
I
initially
started.
G
This
was
like
I'll,
just
be
a
good
little
worker
and
I'll
do
tdd
and
I'll
go
front
to
back,
and
it
like
started
to
churn
a
bit
with
the
internal
state,
and
then
I
was
like
okay,
you
know
what
let's
try
going
backwards,
so
I
like
grabbed
the
otlp
protos
from
metrics.
I
started
implementing
the
encoding.
I
was
like
okay.
This
is
actually
what
the
data
needs
to
look
like
at
some
point.
G
It
started
working
my
way
back
from
there
and
I've
made
a
lot
more
progress
then,
and
now
now
I'm
kind
of
juggling
it's
like
okay,
this
needs
to
be
thread
safe.
It
needs
to
be
atomic,
it
needs
to
not
be
slow.
It
needs
to
not
be
memory
intensive
if
you
have
like
multiple
counters
increasing-
and
it's
like
how
do
I
actually
store
this?
So
that
it's
it's
fast
right,
I
have
something:
that's
kind
of
slow
and
potentially
wrong
right
now,
even
pushed
up.
G
C
F
Slightly,
I
mean
gladly
review,
but
I
was
also
enjoying
the
dig.
He
was
giving
himself
all
right.
D
I
will
say
nicholas,
you
asked
me
hey,
how
can
I
get
involved
and
I'm
like
there's
a
pool
and
I'll
push
you
right
into
the
deep
end.
I
hope
that's.
Okay,.
C
D
D
So
this
one's
a
little
bit
funky
because
in
test
cases,
people
when
they
use
the
faraday
test,
stub
they're
not
required
to
provide
a
url
a
base
url.
So
the
host
name
is
nil
and
that
ends
up
setting
a
netpear
name
attribute
to
nil.
D
What
does
that
do
that
ends
up
calling
the
error
handler?
So
in
our
test
suite,
I
have
the
error
handler
raising
exceptions
when
it
receives
a
invalid
attribute,
because
I
want
to
help
developers
along
and
tell
them
hey
look.
You
know
you've
added
an
attribute
to
this
band
that
doesn't
conform
to
anything
and
we're
not
you
know
we
don't
do
any
checking.
So
I
only
do
this
in
our
test
suite
in
development
mode,
I'm
not
in
production.
D
Obviously,
and
after
updating
recently
that
I
pulled
in
this
edition
of
the
net
peer
name
attribute
and
that
ran
into
an
issue
now.
I
know
some
people
are
sensitive
about
saying
like
well.
If
things
are
set
up
in
test
cases,
they
shouldn't
be
set
up
like
that
and
we
shouldn't
account
for
them
in
production
code,
and
so
that's
something
I
wanted
to.
D
You
know
put
in
front
of
people's
minds
and
then,
in
addition
to
that,
I
would
like
it
if
we
can
extend
as
part
of
like
our
base
sort
of
testing.
Behavior
is
to
have
the
error.
Handlers
raise
an
exception
as
part
of
our
test
suite,
and
then
that
way
we
can
catch
cases
like
this,
like
we
had
this
problem.
D
This
case
also
occurred
with
one
of
the
messaging
attributes,
and
I
think
that
there
was
a
a
fix
for
it
where
sometimes
the
messaging
attribute
was
not
a
string,
but
rather
some
other
ruby
object
when
it
was
getting
pulled
out
of
like
ruby
one
of
the
I'm
gonna
say
it
was
ruby
kafka,
and
so
it
would
constantly
error
out
it
would
it
would
spew
a
bunch
of
error
messages
saying
this
attribute
is
invalid
in
production.
D
G
I
was
just
to
say
yeah.
We
we
introduced
it.
We
didn't
make
it
mandatory,
like
for
people
at
shopify
to
use
the
the
raising
error
handler
and
test
like
we
made
it
available
because,
like
the
very
diligent
eager
good
app
owners,
did
it
we're
doing
it
anyway,
so
we
just
gave
them
a
helper.
We
have
like
a
hand,
still
kind
of
on
the
fence
about
if
I
want
to
like
just
turn
it
on
by
default.
G
But
I
agree
because,
like
we
have
our
log
level
set
to
error
by
default,
the
idea
is
like
people
like,
oh
open,
telemetry,
spitting
out
all
these
errors
and
it's
like
yeah.
That's
because
they're
actionable
go
fix
them,
but
to
kind
of
like
look
at
faraday
the
what
you're
mentioning
does
this
error
only
occur
in
like
people's
setup
tests,
or
is
this
like
running
in
production
that
they're
getting
this.
D
I've
only
seen
it
in
test
cases,
because
in
production
everybody
provides
the
base
url
at
least
at
github.
I
don't
know
what
other
people
do
like.
I
haven't
looked
closely
enough
to
say:
hey
look,
faraday
doesn't
require
base
url
at
all,
and
you
pass
in
the
entire
url
as
part
of
like
the
connections
like
http
request
methods,
and
I
don't
know
like
I
haven't,
dug
into
it
enough
to
know.
G
Let
you
out
of
the
question
if
it
would
break
in
production,
because
it's
just
like
a
not
an
accurate
test
like
in
the
sense
that
it
doesn't
like
reflect
a
working
thing
whenever
I've
encountered
that
with
like
internal
app
owners
like
oh,
my
tests
are
showing
these
arrows
on
open
telemetry
and
it's
like
well,
your
test
doesn't
make
sense
like
this
would
never
work,
and
our
instrumentation
is
identifying
that
and
it's
like,
even
if
it's
doing
it
in
a
roundabout
way
by
saying
like
hey
this
value,
that
should
be
there
isn't
and
you're
getting
error.
G
It's
like
I,
I
don't
know.
If
it's
I
have
the
fortune
of
being
able
to
be
heavy-handed.
Would
those
people
be
like
just
take.
G
G
But
no
it's
a
really
good
point.
That's
exactly
it!
That's
like,
I
think
enough
to
like
change
like
put
me
in
favor
of
like
bringing
in
a
change
like
that.
It's,
like
here's,
a
valid
case.
Okay,
perfect.
Like
then,
there's
no
contention
here,
I'm.
D
I'll
set
up
a
separate
pr
for
that
in
a
sense
like
the
the
perhaps
the
hotel
test
helper
should
have
a
a
default
sdk
configurator
or
something
and
then
have
the
hotel
test
up
or
just
be
like
set
it
up.
And
then
it
sets
up
that
global
exporter
and
spam
processor
blah
blah
blah.
But
I
shall
do
that
in
a
follow-up
activity.
G
Yeah
in
the
test
helper,
I
was
like
trying
to
come
up
with
like
oh
should
we
just
have
this
like
auto
configurator
that
we
use
for
tests,
and
I
was
like
what
would
I
put
in
it,
and
I
didn't
have
an
answer
to
that.
So
I
was
like.
I
guess
we
don't
need
one,
but
now
it
makes
sense
to
potentially
have
one
so
cool
a
lot
of
good
things
coming
out
of
this
pr.
D
G
Honest,
that's
something
internal.
We
have
a
we
fixed
it
today,
yeah
we
monkey
patch,
the
tracer
provider,
so
that
if
you
have
a
non-recording
span,
instead
of
it
just
like
kind
of
having
this
dummy
span
in
there,
the
non-recording
span
will
use
the
parents
fan
id.
So,
if
you
have
like
say
like
a
server
span
that
is
valid
and
then
you
have
like
100
non-recording
spans
and
then
at
the
very
bottom
of
your
your
graph,
you
have
like
a
recording
span.
G
Its
parents
fan
id
will
be
the
server
because
they'll
all
just
point
out
and
it's
to
preserve
the
structure
of
a
trace.
If
we
omit
some
internal
spans.
D
D
For
sure,
speaking
of
getting
their
money's
worth,
rob
you
what
happened
to
the
carry-on
baggage
or
the
carry-on
implementation.
E
Yeah
it's
a
little
embarrassing,
but
where
would
you
want
it
like
it
shouldn't
go
in
this
in
my
pocket.
E
As
an
example
of,
if
you
would
like
to
make
your
stuff
a
mess,
here
is
an
option
for
you
to
contrib.
E
Yeah,
there's
the
prototype
and
then
there's
the
perennial
problem
of
for
those
of
you.
Just
joining
us
carry
on
is
an
implementation
of
baggage
that
doesn't
propagate
it's
like
a
in
process.
I
would,
I
would
like
to
have
a
data
structure
that
I
can
shovel
keys
and
values
into,
and
then,
if
I
also
add
a
processor
that
looks
at
those
values
and
jams
them
on
every
span,
you
can
get
something
that
looks
you
can
the
phrase
that
was
in
chat
earlier,
you
can
increase.
E
The
cardinality
of
your
child's
pants
carry-on
is
an
attempt
to
work
around
the
fact
that
you
can't
there's
there's
nothing
to
within
hotel
to
prevent
you
from
propagating
all
of
these
keys
and
values
that
you've
set
on
baggage.
Every
ongoing
http
request
is
going
to
have
that
repeated
in
the
headers
and
like
people
are
like.
E
Currently,
the
spec
says
yep,
and
so
that's
a
problem
that
you
can
solve
at,
like
your
network
boundaries
with
outgoing
proxies
that
strip
a
header
out
or
something
based
on,
allow
lists
or
denials
so
carry
on
was
like
okay,
you
would
like
to
be
able
to
you.
Have
some
information
from
the
parent
context
you'd
like
to
put
that
information
on
all
of
its
child
spans?
But
you
don't
want
it
to
propagate
it's
very
off-script.
F
E
Okay,
I
will
read
this
discussion
and
either
and
add
to
it
and
maybe
also
put
carry
on
somewhere,
that
it
can
be
seen
publicly
and
opinions
formed
about
it.
G
G
G
Yeah
exactly
like
that's,
why,
like
all
of
our
instrumentation's
gonna,
go
there
the
resource
detectors
that
make
sense
if
we
have
any
like
special
propagators
like
carry
on?
Why
not?
It
seems
like
a
good
idea.
D
A
D
D
Okay,
well,
maybe
next
time
all
right.
Well,
it
is
12
noon.
My
time,
which
means
I
should
need
something.
C
Cool
yeah,
thanks
for
coming
everybody
we'll
see
you
next
week
and
follow
up
with
conversations
online,
otherwise,.