►
From YouTube: 2022-08-23 meeting
Description
cncf-opentelemetry meeting-2's Personal Meeting Room
B
Thank
you.
No
actually,
a
couple
weeks
ago
now
I
got
hit
with
an
escrima
stick
in
the
face
during
one
of
my
martial
arts
classes.
So
everything
is
fine
and
I
don't
know
I
think
everything's
like
90
better
now,
but
there's
still
some
like
shades
of
yellow
and
purple
around
my
my
eye.
A
E
We
should
use
use
your
your
new
image
to
get
specific
stuff
merged.
You
know,
just
like
just
you
know,
all
right
logs
are
we're
doing
1.0
for
logs,
because
matt
says
so.
You
know.
I
think
that
could
be
your
new
role.
Think
about
it.
B
D
I
have
been
here
specifically,
I
just
been
trying
to
get
familiar
with
the
old
hotel
community.
In
general,
I've
seen
rob
before
nice.
B
B
I
guess
we'd
maybe
do
like
a
quick
round
of
intros
so
that
everybody's
familiar
yeah,
so
my
name's
matt
I
work
with
lightsab.
I've
been
around
hotel
ruby
since
the
early
days.
I
mainly
run
this
meeting
and
tell
people
what
happened
at
the
spike
saying,
but
I
try
to
be
useful
when
I
can.
A
America
you
work
with
sam
and
shopify
and
meeting
a
sandwich.
I
apologize
for
this.
D
E
I
can
round
out
shopify
I'm
sam.
I
work
at
shopify
and
I
think
that
nicholas
unmuted,
so
nicholas
can
go
next.
C
Thank
you
sam.
I
thought
we
should
start
doing
that.
I'm
nick
I
work
at
a
company
called
zappy
who's,
lesser
known
than
shopify
it's
about
as
good
as
it
gets
rob.
D
Yeah,
I
yes,
I'm
alex,
and
yes,
I
guess
well
yeah
I
used
to
be
an
internet
honeycomb,
and
I
guess
I
I'll
just
I
guess
I
just
described
myself
now
as
unaffiliated.
I
guess
I
don't
know
I'm
just
here
for
the
hotel
open
source
community.
I
guess
that's
something.
D
B
Fine
we're
we're
happy
to
have
self-affiliated
folks
and
yeah.
There
is.
There
is
a
meeting
doc
if
you've
added
your
name,
somebody's
added
your
name,
but
if
you
haven't
seen
it,
it's
here
feel
free
to
add
items
to
the
agenda
and
yeah
definitely
feel
free
to
stop
by
weekly,
there's,
usually
a
handful
of
people
here.
B
So
I
usually
run
through
a
quick
recap
of
the
spec
sig.
It's
a
long,
boring
meeting
that
happens
before
this
one.
So
I
try
to
condense
it
down
and
sam
tries
to
keep
me
honest,
so
I'll
go
ahead
and
share
my.
B
B
B
Where
yeah
some
metrics
there's
a
a
difference
of
opinion
about
what
should
be
done
about
some
metrics
network
metrics
and
I
think,
like
disk
metrics,
the
two
ways
that
people
are
talking
about.
Recording
them
is
having
like
a
network.io
metric
with
a
direction
attribute
on
it
and
the
direction
would
be
like
sent
or
received,
or
you
could
split
that
into
two
metrics
and
be
like
network.io,
that
sent
and
network
that
io
dot
received,
and
there
are
pros
and
cons
to
both
approaches.
B
There
was
a
spec
change
to
say
that
you
should
split
them,
but
that
went
through
bogdan
was
kind
of
busy
through
a
lot
of
july.
I
think
he
was
on
vacation
and
when
he
came
back
he
saw
this
and
was
not
super
happy.
So
I
think
the
the
debate
continues,
but
this
was
kind
of
holding
up
the
spec
release.
They
decided
at
least
to
revert
this
recommendation
to
split
those
metrics
for
now
to
continue
the
bike
shed
to
actually
figure
out
what
we
want
to
do
so.
B
Moving
on
so.
B
B
B
B
B
Does
anybody
know
how
our
environment
variable
interpretation
works.
B
B
There
was
this
other
question.
It
was
really
mainly,
I
think,
dealing
with
the
java
repo
and
I
think
the
specs
especially
explicitly
says
you
should
not
distribute
the
aws
propagator
in
like
the
core
main
repo.
Basically,
it
should
be
a
contrib,
but
I
think
java
is
distributing
it
out
of
their
main
repo
and
they
were
wondering
like.
B
Should
they
try
to
change
the
specs?
Should
they
try
to
change
what
they're
actually
doing,
and
I
think
the
tldr
is
it's
a
java
problem
and
they
will
come
up
with
some
sane
solution
to
it
and
yeah.
B
About
where
those
exporters
should
reside,
as
it
was
kind
of
rekindled,
some
people,
like
I
don't
know
somebody
from
google
was
like
well.
If
we
did
this
for
aws,
maybe
we
should
also
do
it
for
google
and
blah
blah
blah,
but
even
the
googler
was
saying,
but
we
should
actually
just
switch
to
w3c
all
of
us
and
then
we
have
no
problems.
So
I
think
most
people
agreed
with
that
and
I
think
that's
the
real
answer
and
solution.
B
A
Wait
one
what
so,
what
was
the
the
ideas
these
ought
to
live
and
contribute,
though
right.
F
F
A
Yeah
I
got
asked
it's
funny
when
I
did
something
similar.
I
had
a
exporter
for
datadog
way
way
back
before.
That
was
a
terrible
approach.
I
got
kicked
out
of
one
of
the
like
this
this
la
this
here
I
got
kicked
out
of
this
sig
not
kicked
out
like
I
wasn't.
They
said
just
host
it
on
your
own
and
add
it
to
the
registry,
but
then
python
was
like
contrib
shove
it
in
there
and
there
is
a
meaningful
difference
in
adoption
if
you
can
get
into
contrib
versus.
A
F
F
E
A
B
Yeah,
I
think
I
think,
that's
reasonable.
I
vaguely
remember
us
suggesting
that
you
at
the
time
kept
that
in
the
datadog
repo,
and
I
think
that
is
where
a
lot
of
things
reside
for
for
a
lot.
B
D
B
The
other
thing
that
we
were
talking
about
is
the
the
code
button
suggestion
to
emit
a
or
to
send
a
user
agent
header
from
hotel
exporters,
and
I
think
it
has
enough
approvals.
So
it's
likely
to
merge
soon.
F
So
here's
the
thing
we
asked
for
and
I'll
present
it
to
this
brain
trust
we
we
had
desired
to
put
into
this
user
agent
string.
In
addition
to
like
the
example,
is
go
because
the
go
hotel
exporter
just
says:
hey
I'm
using
a
grpc
library
from
go
and
it's
like
are
you?
Are
you
the
open?
Telemetry
agent?
Are
you
what
are
like?
F
There's
no
differentiating
other
grpc
centers
like
there's,
no
differentiating
go
sdk
directly,
sending
to
a
receiver
versus
passing
through
a
collector,
except
now
the
collector
adds
itself
to
the
user
agent,
because
we
submitted
that
pr
to
the
user
agent.
We
were
asking
for
the
otlp
proto
version,
because
receivers
who
need
to
receive
different
versions
for
reasons
like
amazon's
collector,
is
still
running
0.70
of
the
proto
otlp
proto.
F
F
B
Okay,
so
I
will
at
least
say
that
there
was
some
discussion
about
the
version
number
a
little
bit.
I
will
try
to
dig
this
stuff
up
and
pass
it
along
cool
and
yeah
just
to
wrap
this
up.
The
last
thing
was
just
a
long
long
long
bike
chat
about
this
short
name
attribute
that
they
are
suggesting
to
add.
This
is
to
prefix
prometheus
metrics,
but
the
bike
shed
was
worthy.
B
I
think-
and
ultimately
I
think
the
bike
chat
is
like:
isn't
the
instrumentation
scope,
the
short
name
and
then
they're
like
well,
it's
actually
too
long
and
then
like?
What
would
you
do
with
a
short
name
other
than
prefix
prometheus
metrics?
And
then
people
were
recommending
a
lot
of
things.
They
would
use
it
as
a
join
key.
They
would
use
a
lot
of
other
things,
so
I
can't
really
use
the
short
name
as
a
join
key.
B
You
would
have
to
use
the
full
name,
because
the
full
name
gives
you
uniqueness
and
that's
why
it's
so
long
to
begin
with,
and
so
I
think
there
was
no
resolution,
but
out
of
that
discussion
I
think
the
thing
that
people
realize
is
like
this
just
seems
to
be
like
another
hack
on
top
of
scope,
name,
so
the
I
think
people
are
still
kind
of
resisting
this
change
overall
and
trying
to
figure
out
like
what
can
we
do
to
scope
name
to
make
that
the
thing
that
you
can
use
for
all
this
stuff?
B
D
Not
yeah,
but
I
missed
one
sig
meeting
and
people
were
apparently
beating
each
other
up
with
sticks
or
something.
E
B
All
right
so
moving
on
are
there
any.
D
B
A
Oh
I
had
follow
up
last
week
from
last
week.
I
was
supposed
to
go
and
see
what
needed
to
be
released.
I
didn't
do
any
of
that.
So
that's!
I
should
probably
do
that
this
week.
Sorry.
D
A
B
So
the
tl
dr
is,
we
are
now
checking
in
to
see
if
we
should
release
something
we
did
this
last
week,
and
the
action
item
was
to
was
to
execute
the
action
item,
which
did
not
happen
so
today,
are
we
basically
carrying
over
the
action
item?
Are
we
going
to
see
if
anything
should
be
released,
or
are
we
going
to
punch
this
until
kind
of
next
week.
A
Yeah
I'll
look
into
it
this
week.
I
say
out
loud
so
that
that's
so
you
know
it's
not
obligated
to
yeah.
I
I
do
still
think
there's
some
pr's
I'd
like
to
see
there's
a
couple
pr's
in
there
that
we
ought
to
really
get
merged,
but
yeah
such
as
nicholases
and
then
I
think,
a
couple
things
that
are
just
approved
by
bunchbook,
so
yeah,
just
gotta
do
my
job
and
I
haven't
and
yeah.
You
know.
B
B
Yeah
that's
clear:
if
we
kind
of
discuss
this
and
as
long
as
there's
nothing
like
pressing,
then
I
think
yeah
karen
carrying
this
over
through
the
week
makes
sense.
A
D
A
There's
like
some
pr's
that
came
through
the
pipe
one
like
nicholas's
is
probably
worth
I
don't
know.
What
did
I
have
to
pay
attention
to?
It
seemed
good,
then
the
mongodb
thing
is
like
pretty
minor
but
worth
but
open
to
bike
shedding.
If
people
just
really
want
to
kill
30
minutes,
but
yeah,
maybe
nicholas,
do
you
want
to?
Maybe
we
should
focus
on
his.
C
C
But
basically,
if
you
do
not
have
a
way
of
doing
any
sort
of
flushing
in
the
child
process,
then
oh,
let
me
start
from
the
beginning.
So
the
way
that
rescue
works
is
a
worker
will
pick
up
jobs
from
a
rescue
and
then
just
to
make
sure
that
that
worker
doesn't
die
a
sad
death
like
during
ooming
or
something
like
that,
while
performing
the
job
it
spawns
off
a
child
process
which
can
be
as
volatile
as
it
wants
once
that
child
process
is
done,
processing
or
performing
the
job.
C
It
then
terminates
itself
and
if
you
have
a
relatively
vanilla
rescue
job-
and
you
don't
have
many
after
or
round
hooks,
then
that
job
kills
itself
or
the
child
process
kills
itself
before
the
ozal
span.
Processor
can
be
flushed,
so
we
were
running
into
issues
where
I
didn't
notice
any
spans
coming
through
for
the
process
action
and
the
idea
is
to
basically
just
adopt
what
was
done
for
active
job
where,
if
it
was
configured
to
true,
then
we
force
flush.
The
span
processes
before
the
child
process
exits.
F
This
comes
up
with
lambdas,
too
anybody
using
hotel
or
anything
that
ships
things
in
all
in
in
other
threads
and
cues
processed
by
other
threads.
When
your
execution
environment
is
like
all
right,
I'm
done
with
main
thread
the
the
any
sort
of
batching
it
has.
The
risk
of
getting
chucked
as
the
execution
of
arm
stands
down.
F
So
I'm
a
fan
of
this
option,
but
also
there's
like
this
larger
pattern
of
like
how
can
we
teach
hotel
batch
span,
processors
to
like
and
then
anything
any
execution
using
those
things
in
execution
environments
that
could
suddenly
shut
down.
F
C
The
only
real
thing
is
that
it's
got
two
approvals
and
I
just
noticed
that
ruby
3.1
is
falling
over.
I'm
not
super
familiar
with
the
way
that
the
specs
are
being
done,
but
it
does
seem
like
there's
a
bit
of
a
heisen
spec
in
that
sometimes
the
rescue
instance
is
instrumented,
with
the
config
being
false
or
true,
and
depending
on
that
order,
then
the
mini
test
mini
cube
test
mock
falls
over
because
the
mark's
built
in
a
way
where
it
expects
it
to
be
run
in
order.
C
It's
kind
of
that
yeah,
but
it's
not
the
two
specs
themselves
that
need
to
be
run
in
order.
It's
the
way
the
entire
spec
files
run
yeah
so
based
on
seeding
and
randomization,
which
I
assume
is
the
way
that
these
specs
are
being
run
not
in
order.
Then
it's
running
it's
in
a
heisenspec.
So
if
you
try
recreate
this
on
local
it'll
work,
six
times
break
on
the
seventh
and
so.
F
C
Yes,
I'm
not
sure
if
this
is
also
a
heisen
spec
in
the
active
job
side
of
things.
I
didn't
go,
look
at
flaky
specs
and
circle,
but
I
basically
just
took
andrew's
spec
and
rewrote
it
because
it's
the
same
instrumentation,
but
the
spec
file
itself
is
slightly
shaped
slightly
differently.
So
I'll
just
look
to
make
it
a
bit
more
clean
and
reliable.
C
I
think
the
only
thing
from
my
side
was
I'm
not
so
sure
where
we
land
on
the
pattern,
but
I
was
wondering
if
we
should
keep
configuration
explainers
as
comments
code,
or
we
should
also
have
those
added
to
the
base
instrument
based
instrumentation
readmes.
I
noticed
that
some
instrumentations
have
the
options
defined
and
explained
with
the
default
configuration
in
the
readme.
C
So,
like
rest,
excuse
me,
but
others
don't
and
I'm
thinking
in
terms
of
like
a
new
adapter,
it
might
be
easier
to
have
that
present
in
the
base
readme
for
the
instrumentation
versus
having
to
like
navigate
the
the
way
the
instrumentation
is
kind
of
layered,
deeper
or
nested
deeper
into
folders,
and
then
find
the
code
and
see
it
there.
B
B
I
also
realize
that
the
readme
can
easily
get
out
of
date
get
out
of
sync
with
stuff,
so
I
think
that
ends
up
being
a
consideration,
but
I
think
maybe
we
should
have
a
yeah.
We
should
see
what
the
group
things
and
see
if
we
have
a
way
to
handle
this,
because
we
have
many.
F
My
my
initial
reaction
is
a:
why
not
both
situation
where
I,
like
the
I,
like
the
ideal
of
the
code,
docs
turned
into
browsable
documentation,
so
like
the
stuff
that
shows
up
at
rubydoc
dot
info
right,
I
think
we
should
keep
we
we
shouldn't
remove
it
as
code
documentation.
F
There
could
be
some
value
at
higher
up
in
the
class
hierarchy
like
at
your
entry
point
into
this
instrumentation
class.
You
can
point
to
the
lower
level
stuff
like
see
this
see
the
documentation
on
this
class
or
this
class
and
method
on
usage
so
that
when
you're
at
ruby.info
it
drives
it.
It
links
you
deeper
down
where
the
details
are,
and
then
the
readme
for
the
gem
can
point
you
there.
F
F
Like
so,
the
readme
for
the
gem
could
have
like
the
happy
path,
so
you're
gonna,
add
this
in
and
you
know,
but
for
configuration
details
see
the
the
api
documentation
that
was
generated
from
the
code
document
code
comments
is
that
it's
not
super
convenient
all
in
the
readme,
but
it
also
keeps
the
readme
from
one
divert
like
a
human
human
doesn't
have
to
keep
them
from
diverging.
F
B
The
more
I
think
about
it,
the
more
I
feel
like
having
a
config
heading
on
these
readmes
that
just
have
a
table
of
the
options
and
what
they
do.
I
think
it
makes
a
lot
of
sense
and
will
probably
save
people
a
lot
of
time
in
the
long
haul.
If
we
feel
like
we
can
be
reasonably
honest
about
keeping
those
up
today.
B
F
F
B
F
I
think
nick's
point
about
it
being
buried
is
like
you
have
to
go
and
navigate
the
gem
internal
structure
to
find
details
is
a
is
a
flaw
that
we
should
fix.
Yeah.
C
Yeah,
but
I
think,
like
I,
like
the
r-doc
and
not
read
me
just
because
then
you're
not
maintaining
two
spaces,
but
I
think
that
only
works
if
you
have
some
sort
of
my
brother's
dog,
so
sad
that
he's
gone.
C
If
you
have
some
sort
of
like
established
link
on
hotel
contra
base,
read
me
it's
like
any
options
where
any
instrumentations
can
be
found
in
this
path
or
this
pattern,
or
something
like
that,
just
to
kind
of
make
people
aware
of
it
being
there,
because
even
if
you're
doing
like
instrumentations
all,
I
guess
there
are
examples
of
configurable
options
there.
Just
helping
people
find
out
where
the
information
about
those
configs
are
and
then
having
that
be
consistent,
because
my
main,
my
main
issue
with
that
was
just
that.
F
All
right
does
anybody
disagree
that,
as
we
write
code
comments,
we
should
follow
some
sort
of
formatting
mechanism
that
makes
the
generated
api
docs
better.
I
think
I'm
pro
doing
that
as
we
start
touching
these
things,
we
should
make
our
generated
documentation
better
and
then
the
better
that
gets
the
more
we
can
link
to
it.
F
B
I
think
that
makes
sense,
I
think,
having
an
example
helps
and
if
we
can
update
our
code
to
that
example,
and
I
think
it'll.
E
B
Yes,
indeed,
so
so
yeah,
I'm
not
sure
if
this
is
like
issue
worthy
to
kind
of
come
up
with
a
process
to
make
this
a
little
bit
more
uniform
around
things
but
yeah.
It
sounds
like
in
theory.
B
If,
for
the
most
part,
we
have
these
comments
in
our
code,
they
might
not
be
structured
in
a
way
that
degenerated
docs
are
as
useful
as
they
could
be,
so
the
next
step
would
be
to
figure
out
how
to
properly
annotate
the
docs
so
that
we
can
generate
the
thing
that
we
want
and
then
just
kind
of
do
it
everywhere.
B
B
C
B
E
Yeah,
this
is
not
particularly
urgent
at
all,
so
if
there's
anything
else
that
actually
matters,
I
would
say
definitely
talk
about
those
first,
because
this
is
just
kind
of
like
an
idle
thought.
E
E
So
I
was
looking
into
basically
annotating
or
andrew,
and
I
were
looking
into
annotating
bug
snag
reports
with
a
trace
id
and
a
span
id
bugs
nag
is
for
error
collection
if
you're
not
familiar
it's
like
century,
but
different
or
kind
of
it
doesn't
matter
and
there's
this
kind
of
interesting
problem
where
our
libraries
in
span
helper
function,
like
yields,
a
block
and
it
basically
like
catches,
any
exception
that
gets
raised
during
the
spans
processing
and
then
it
like
annotates.
E
That
span
with
the
exception,
details
and
bug,
snag
and
then
and
then
it
finished,
and
then
it
ends
the
span
is
finished
and
bug
snag
kind
of
has
a
similar
tri-catch
mechanism,
but
it
happens
outside
of
the
hotel
span
like
bug.
E
Snag,
is
usually
registered
as
like
middleware
in
some
sort
of
processing,
like
whatever
in
in
sorry,
you
registered
middleware
for
like
a
rack
application
or
like
in
rescue
or
sidekick
or
whatever,
like
there's
some
kind
of
middleware
whatever,
so
that
kind
of
like
by
the
time
you
can
basically
write.
Custom
bug,
snag
middlewares,
which
run
you
know
when
bug
snag
gets
the
exception.
E
But
by
that
time
the
open,
telemetry
context,
like
the
span
that
contains
the
error,
has
basically
been
popped
off
of
the
context
stack
so
you're
sort
of
in
like
one
layer
up
in
context
or,
however
many
layers
up
in
in
the
context
so
kind
of
like
trying
to
figure
out.
E
Is
there
a
way
to
like
you
can
do
a
very
rough,
not
very
somewhat
rough
thing
where
it's
just
like
hey
like
if
there
is
a
trace
id,
let's
just
assume
that
there's
one
trace
for
you
know
this
rescue
job
or
this
sidekick
job,
or
this
rack
request.
E
Let's
assume
that
everything
belongs
to
a
single
trace,
that's
a
safe
assumption
and
you
can
just
grab
the
trace
id
off
of,
like
whatever
span
is
in
the
current
context,
and
it's
like
that's
close
enough,
but
I
was
curious
if
folks
had
any
thoughts
of
like
a
on
like
an
hotel
compliant
weight
like
yeah,
you
got,
it
nicholas.
Has
it
in
the
chat
and
has
a
like
an
like
a
hotel
compliance
way
to
get
the
span
id
without
doing
something
super
gross?
I
mean,
like.
E
I
think
it's
just
going
to
be
gross.
If
you
want
to
get
the
span
id
to
attach
to
a
sentry
at
a
bug,
snag
air
report,
but
anyway
did
that.
Does
that
problem
make
sense
to
people
you
can
you
can
log
off?
If
you
want
it's
just
something
that
I've
been
thinking
about.
B
Yeah,
I
think
the
problem
definitely
makes
sense.
It's
like
you
want.
You
want
a
span
id
correlated
with
your
error,
but
the
span
is
long
gone
where
the
error
created
by
the
time
yeah
the
bug
snag.
Stuff
is
running.
So
it's
kind
of
one
of
these
situations.
B
B
C
Just
to
kind
of
add,
because
we
use
sentry
and
from
our
perspective
it's
actually
something
that
one
of
my
teammates
is
trying
to
solve
this
week
and
you
get
the
same
problem
where
you
have
this
exception
being
raised
and
it
gets
handled
now.
We
want
to
start
adding
it's
easy
to
add
the
trace
id,
but
we
have
these
this
issue,
where
the
handling
of
sending
that
event
to
sentry
is
asynchronous.
C
So
in
that
forked
process,
when
you
try
to
do
current
span,
there
is
no
current
span
and
you
can't
even
get
the
trace
id
so
yeah.
Something.
F
C
A
E
Yeah
I
mean
like
right:
you
can
just
shove
it
somewhere.
Like
I
mean
that
was
my
impulse
as
like
a
complete
hack
that
was
doing
devops
for
the
past
five
years.
I
was
just
like
yeah,
like
just
that.
Yeah
put
it
on
the
thread
just
some
thread.
Local
storage-
don't
be,
don't
be
a
baby,
but
that
is
not.
But
that
is
what
rack.
That
is
what
we
do
for
rack.
E
I
mean
rack,
there's
like
a
terminating
exception,
like
part
like
variable
on
the
environment,
on
the
rack-
and
I
don't
know
if
that's
internal
to.
A
My
practical
advice,
not
speaking
as
your
co-worker,
is
I'm
pretty
sure
bug
snag's
been
showing
up
to
a
bunch
of
inflammatory
stuff
recently
and
seems
interested
in
compatibility
with
open
telemetry
so
like
it
would
be
dumb
to
roll
some
hacky,
whatever
not
so
happy
thing
goodbye,
andrew
and
then
have
bug
snack
come
out
with
like
official
support
for
it
some
other
way,
although
their
official
support
sounds
also
like
it
would
have
to
be
pretty
you
know,
hacky
so
yeah,
I
would
maybe
touch
base
with
the
who
I
can
track
down
baby.
C
Now,
you're
just
waving
your
we're
a
big
company,
we
can
solve
this
problem.
Therefore,
hotel
doesn't
have
to
solve
this
problem
which
might
affect
all
the
other
people
using
different
vendors
and
aren't
a
big
company.
F
C
F
But
actually
that's
a
good
idea
connect
with
the
the
bug
snaggers
that
are
getting
involved
in
hotel
and
see
if
there's
a.
F
A
E
That
this
confirms
kind
of
I'm
I'm
glad
that
this
is
a
common
problem.
I
mean
not
glad,
but
it's
nice
to
know
that
we're
not
alone
in
this,
because
I
was
like
no
biggie
I'll.
Just
like
add
a
bug,
snag
call
back
add-on
error
hook
and
call
it
a
day
like
nope,
not
gonna
happen
no
nice.
I
joined
this
team
a
little
bit
late
for
all
the
low
hanging.
Fruit
is
what
is
what's
happened
so.
B
B
Things
going
on
right
now
we
have
the
in.
So
I
think
if
we
want
to
understand
the
code
path,
sam
was
talking
about
it's
kind
of
in
span.
We
end
up
calling
this
spam
record
exception.
B
B
It
just
makes
an
event
yeah
of
type
exception,
but
I
don't
see
it
running
through
this
error
handler,
which
is
the
other
thing
that
we
were
talking
about.
Yep.
F
B
E
F
E
Variable
set
yeah,
that's
that
is,
if
you're
talking
about
monkey
patching,
that's
yeah.
I.
E
Definitely
does
uses
that
method
they
just
like
mess
with
the
exception
class,
except
that
we
don't
really
give
you.
F
Jam
these
instance
variables
on
an
exception
mechanism.
That's
you'd
have
to.
E
E
F
E
Yeah
agreed
dude,
we
I
mean
everyone
should
know,
but
we've.
F
We've
got
the
you
know,
add
stuff
to
context
mechanism
that
I
that
yeah,
what
what
eric
said,
use
context
directly
to
shove
stuff
into
context,
but
you're
right.
The
russian
doll
nesting
of
context
means
that
the
thing
that
you
annotated,
that
with
yeah,
maybe
disappear.
Much
like
this
band
did.
E
Yeah,
I
can
type
it
up.
It's.
F
F
D
This
is,
this
is
correlating
the
current
span
id
with
a
with.
D
And
that,
and
and
and
that,
because
bug
snag,
isn't
in
the
workflow
of
record
exception.
This
is
why
somebody
was
suggesting
use
record
exception,
decorate.
C
F
I
think
eric's
suggestion
remains
the
most
sensible
which
is
which
is
to
use
bug.
Snacks,
no
bug
snack
people
are
getting
involved
in
hotel
and
maybe
go
talk
to
them
and
see
what
we
could
do
about
getting
the
two
to
dance
together.
F
F
D
In
my
head,
it's
like
bug
snag
should
be
using
the
event
that
it's
been
recorded
when
and
how
bug
snag
should
have
an
exporter.
The
problem
is
that
the
exception
has
been
serialized
at
that
point
into
different
semantic
convention,
attributes
on
the
log
event
or
the
other
the
span
event,
and
so
you
don't
have
like
the
you.
Don't
have
a
way
to
tr.
You
know
to
to
kind
of
like
deal
with
very
long
stack,
traces
right
like,
for
example,
you
full
message
for
like
a
like.
D
F
We
could
choose
to
like
an
option
we
have
and
no
I
don't
know
that
it's
a
good
one
is
in
hotels
in
span
exception,
handling,
modify
the
exception
like
yeah
instance
variable
set
span
id
and
anything.
Then
you
re-raise
that
sucker
and
now
it's
a
instance
variable
on
the
exception
that
anything
you
can
choose
and
handling
the
exception
to
know
the
span
id
that
generated
it.
D
E
D
F
D
It
has
record
exception
and
it
says
in
record
exception.
The
results
of
this
will
be,
you
add,
an
event
that
has
these
attributes
so
essentially
record
exception
is
a
span
processor
helper
function
that
is
on
the
sdk
object,
instead
of
it
being
a
processor
on
the
processor
pipeline
and
the
error
or
exception
thing
that
is
recorded
is
transformed
as
part
of
the
client-side
processing
pipeline
or
that
there's
some
sort
of
way
because,
like
if
you,
if
you
wanted
to
export
these
exceptions,
to
bug
snag
from
the
collector.
E
E
F
E
B
B
D
F
D
My
goodness
alexander,
I'm
sorry
we
didn't
get
to
meet
around
here
learning
you
can
hit
him
up
in
the
recording,
hi
and
nicholas
good
to
see
you
again,
yeah.