►
From YouTube: 2023-01-18 meeting
Description
cncf-opentelemetry meeting-2's Personal Meeting Room
A
A
A
B
B
Please
add
your
name
to
the
attendees
list.
If
you
can
and
if
you
have
any
agent
items
you'd
like
to
discuss,
please
add
them
to
the
agenda.
B
C
C
We
want
to
kind
of
Define
how
instrumentation
should
behave
that
isn't
necessarily
A
semantic
convention,
but
kind
of
formalize
a
little
bit
more.
The
structure
so
like,
for
example,
the
event
to
carrier.
You
want
to
have
that
in
the
spec
somewhere.
B
Right,
thank
you,
yeah.
That
reminds
me
now
yeah,
so
I
think
the
the
question
was
whether
we
wanted
to
try
to
have
that
definition
put
directly
into
the
spec
repo
or
if
you
wanted
to
try
to
Stage
it
into
the
Lambda
repo
as
a
place
where
we
can
quickly
iterate
and
then,
when
things
are
setting
more
formed,
bring
that
to
the
spec
repo.
C
The
matter
of
like
how
to
structure
it
within
the
spec
repo,
because
right
now
the
spec
repos
split
up
along
the
you
know,
axes
of
tracing
metrics
stuff
like
that,
and
there
isn't
neces
I,
mean
I,
think
there's
a
common
folder.
So
maybe
it
would
make
sense
to
put
in
common
or
we
could
have
a
separate
instrumentation
specification.
Folder.
B
All
right,
yeah,
yeah,
so
I
guess
the
the
question
here
is
for
anyone
who
has
opinions
about
where
and
how
we
should
put
the
the
definition
of
how
these
instrumentations
should
behave.
Should
we
should
we
do
that
directly
in
the
Linda
repo?
Should
we
try
to
do
it
in
the
spec
repo
initially
or
later,
and
if
we
do
it
in
the
spec
repo?
B
Where
should
that
live
because,
as
Tyler
mentioned,
it's
kind
of
living
right
now
in
the
race
semantic
conventions,
which
seems
a
little
bit
weird,
to
have
description
of
where
the
they're
description
of
how
the
the
land
instrumentation
should
function
in
the
trace
semantic
conventions.
B
So
I
don't
know
that
we
need
to
have
a
resolution
on
that,
but
if
people
have
thoughts
feel
free
to
chime
in
on
any
of
the
the
PRS
that
are
made
or
in
the
slack
Channel
yeah
I
think
that's
about
all.
We
need
to
say
about
that.
One.
B
Well,
I
guess
we
could
take
one
step
back
and
and
ask
one
of
the
things
that
I
had
suggested
yesterday
and
I
think
there
seemed
to
be
General
agreement
on
Tyler
correction.
If
you
have
a
different
opinion
on
this,
though,
is
that
we
should
have
some
specification
of
standard
elements
of
the
land
base
rotation
across
multiple
languages.
So
you
know
there
should
be
these
behaviors.
There
should
be
these
controls
for
for
the
user
who's
trying
to
use
this
instrumentation,
and
they
should
be.
You
know
consistently
named
and
structured.
C
Yeah
I
definitely
think
that
it
should
be
a
should
and
not
a
must
level
of
requirement,
but
defining
that
at
a
high
level
and
expecting
other
languages
other
instrumentations
to
follow
that
unless
they
have
a
really
good
reason,
I
think
makes
sense.
B
Okay
to
move
on
yesterday,
we
discussed
a
fair
bit
of
how
how
to
determine
from
Atlanta
events,
which
carrier
should
be
used
for
context
propagation,
and
we
did
a
brief
survey
of
how
some
of
the
the
languages
are
currently
doing
it.
B
Tyler
I
think
you
said
that
the
Java
implementation
was
a
bit
more
complex
than
you
had
initially
thought
yesterday.
Can
you
give
us
an
update
on
yes.
C
C
There
is
some
cases
where
I
think
it's
to
some
degree
broken
like
it's
basically
taking
in
terms
of
a
lot
of
the
the
the
event
types
it's
it
seems
like
it's
ignoring
the
the
in
the
event
and
just
passing
in
no
op
or
an
empty,
an
empty
carrier,
and
so
I,
don't
know
why
it
was
done
that
way
initially.
C
But
in
my
mind
that
feels
in
some
ways
broken
so
I
think
there's
definitely
some
improvements.
Can
that
could
be
made
there,
and
you
know
migrating
to
this
event
to
carrier
kind
of
structure
seems
like
a
a
good
point
to
do
that.
B
And
then
to
back
up
slightly
for
for
those
who
were
in
the
discussion
yesterday,
basically,
we
were
proposing
that
there'd
be
a
a
hook
or
a
user
provided
callback
that
the
instrumentation
would
use
that
takes
in
raw
events
and
returns
carrier
that
can
be
used
with
context
propagators,
and
so
it
would
be
the
user's
responsibility
to
get
a
carrier
out
of
the
event,
and
then
it
would
be
the
instrumentation's
responsibility
to
apply
the
context
propagator
to
extract
a
context
from
that
carrier
that
the
users
down
and
we
found
that
there
were
a
few
different
ways
of
doing
it.
B
Go
has
event
a
carrier
as
I
just
described.
Java
has
deserialization
prior
to
handling
handing
it
to
the
function
by
overloading
types
that
I
think
it's
how
Tyler
described
it
and
then
JavaScript
and
python
have
a
hook
for
event
to
context
which
combines
extracting
the
carrier
from
the
event
and
extracting
the
context
from
the
carrier.
B
Our
suggestion
is
going
to
be
that
we
separate
those
two
things
so
that
getting
the
carrier
out
of
the
event
is
one
step
and
that's
the
responsibility
of
the
application
author
or
the
user
of
the
instrumentation,
and
then
the
extraction
of
the
context
from
that
carrier
is
handled
by
the
instrumentation,
because
at
that
point
it's
always
just
applying
the
context.
Propagator.
B
B
So
it
sounds
like
the
next
step
on
that,
then
is
going
to
be
to
try
to
specify
the
behavior
of
that
event
to
carrier
fallback.
I
can
go
ahead
and
take
an
action
item
to
take
a
first
pass
at
that.
Try
to
provide
some
language
that
we
could
use
to
to
make
that
consistent
across
the
London
or
across
all
of
the
implementations.
B
One
of
the
other
things
we
discussed
yesterday
was
looking
for
the
feature
compatibility
Matrix,
so
we
can
just
see
what
features
all
of
the
implementations
have
presently
and
get
a
sense
of
where
we've
got
consistency,
where
we've
got
Divergence
and
and
what
work
is
ahead
of
us
in
order
to
try
to
reconcile
all
of
the
the
implementations
we
had
volunteers
for
creating
that
Matrix
and
assessing
most
of
the
languages
that
currently
have
implementations.
We
are
looking
for
someone
who
can
do
an
assessment
of.net.
B
Yeah
I
think
I
I
can
think
of
that
stamina.
Formatting
can
do
it.
B
Yeah
I
said
that
someone
for
my
team
can
do
it.
Yeah.
B
I
will
put
your
name
down
and
you
can
do
what's
the
elevation
if
you
like,.
B
Hey
Next,
Step
I'm
interested
are
looking
at
the
the
current
specs
that
that
exists
and
could
use
some
help,
especially
with
regard
to
other
functions
of
service
providers,
then
Lambda,
so
GCT
or
azure
I
believe
they
have
names
for
some
people
to
reach
out
to.
But
if
any
anyone
here
knows,
anyone
who
is
very
familiar
with
gcp
functions
or
Azure
functions
and
would
be
interested
in
participating
in
this
effort
going
to
take
us
an
interdirection
would
be
greatly
appreciated.
D
So
I
mean
all
I
can
add
to
that
I
added
to
the
agenda
as
well.
We
have
now
a
little
bit
of
gcp
documentation
on
the
open,
Telemetry
IO
website
I'm
also
not
an
expert
on
that.
But
but
we
thought
like
having
a
little
bit
there,
maybe
encouraging
people
to
to
look
at
it,
trying
it
out
and
then
maybe
raising
some
some
questions.
D
I
don't
know
but
beyond
it
I
think
there's
an
Hotel
gcp
slack
channel.
So
maybe
we
can
also
ping
people
over
there
and
see
if
someone
wants
to
jump
onto
that,
not
sure
about
a
sure.
C
Yeah
I
think
the
main
thing
that
we're
looking
at
right
now
is,
if
anyone's
interested
in
those
those
vendors
specific
function
as
a
service
products
to
help
us
align
the
terminology
and
the
naming
just
to
make
sure
that
we
don't
have
terms
that
are
conflicting.
Or
you
know
too
specific
to
one
vendor.
In
the
specification.
D
Yeah,
dare
I
cannot
help,
except
maybe
asking
around
if
I
can
find
someone.
C
B
I
think
it's
about,
oh,
we
can
ask
for
on
on
this
at
this
time.
So
thank
you
for
that.
Okay,
async
scenarios.
So
there
were
some
questions
about.
How
do
we
deal
with?
B
What
are
the
parent
contexts
for
async
messaging
scenarios?
I
believe,
do
we
have
any
comments
on
this,
or
is
this
something
we
just
didn't
reach
yesterday,
and
this
didn't
have.
C
So
if
you
look
down
at
the
the
notes,
a
little
bit,
I
did
mention
a
little
bit
of
this
discussion.
Yesterday,
the
the
main
discussion
was
around
in
like
the
messaging
Sig,
we're
talking
about
you
know
the
the
debate
between
parent
child
versus
span
link
relationships
between
message,
producer
and
consumers,
and
that
comes
into
play
pretty
significantly
when
dealing
with
Lambda,
sqs
or
SNS
event
handlers,
which
is
why
it's
relevant
to
this
group.
B
Okay
I
know:
I
I
believe
that
Lambda
and
x-ray
recently
did
some
work
regarding
linking
of
sqs
events.
Lubin.
Were
you
part
of
that
effort,
or
do
you
have
any
any
information
on
that?
That
could
be
useful
for
us
as
we
look
at
how
to
do
that
with
opens
launcher.
B
You
repeat
again:
yes,
the
the
question
is
basically,
what
is
the
the
appropriate
way
to
handle
context,
propagation
and
setting
the
parent
context
when
dealing
with
asynchronous
communication?
So
maybe
you've
got
a
batch
of
sqs
messages
that
come
into
Orlando
function
and
so
you've
got
a
span.
That's
response
that
covers
the
the
work.
That's
handling
that
badge,
but
each
of
the
items
in
that
badge
also
have
some
work
and
each
of
them
may
have
another
context
associated
with
them.
B
E
I,
remember
that
that
that
thing
it
was
that
one
I
was
in
the
discussions.
I
know
a
little
bit.
I
was
not
directly
involved,
so
the
way
that
it
it
we
solved
that
issue
and
I
think
it's
only
soft
for
rescue
guys.
The
I
I
I
miss
the
details.
So
we
do
this
for
for
for
X-ray,
and
it
is
the
Polaris
that
emits
some
kind
of
a
linking.
E
The
parent
of
these
traces
here-
okay,
not
the
these
multiple
traces
that
are
that
are
embedded
in
this
case
event
metadata.
They
all
continue
with
this
other
segment.
That
would
be
the
alarmifications
and
Trace
context
yeah,
so
it
happens
outside
of
what
we
have
in
in
the
it
kept
us
outside
of
yeah.
So
it's
not
it's
not
instrumented
in
the
in
the
applications
that
in
the
Lambda
functions
it's
not.
B
E
So
it's
a
little
bit
more
complicated,
there
is
SKS
service
and
then
there
is
SKS
power
that
is
kind
of
that
is
kind
of
polling
cache
place
and
when
it
gets
some
messages,
depending
on
some
rules
it
with
batch
of
messages
sitting
folks,
along
with
that
that
is
head
of
all
these
messages,
so
the
actual
linking
is
done
by
DSS
polar
that
is
since
between
the
Lambda
execution
and
SKS
itself.
C
So
that's
what
you're
talking
about
is
specifically
how
x-ray
is
going
about
solving
this
particular
problem.
A
C
So
what
we're
trying
to
get
at
is
what
what's
the
best
way
to
handle
this
from
an
instrumentation
perspective
like
do
we
like
having
span
links,
it
seems
like
span
links,
are
not
particularly
well
supported
by
most
vendors
to
be
honest
or
most
systems,
and
it
has
obviously
you
know,
sampling
issues
around
the
the
fact
that
their
independent
traces
and
thus
generally
have
independent
sampling
decisions,
and
so
you
have
a
with
any
amount
of
sampling
taking
place.
C
C
So
in
some
systems
that's
I
mean
in
some
architectural
designs.
That's
probably
okay,
but
I
I
know
for
sure
that
in
some
architectures
having
it
be,
a
parent-child
relationship
is
going
to
result
in
a
much
better
experience.
E
The
the
problem
here
is
that
you
have
okay
from
from
yeah
I
agree
that
spellings
or
linking
segments
are
called
in
x-ray
is
sadly,
and
the
problem
is
that
you
have
to
or
who
you
are
creating
a
one
new
case
context
for
the
Lambda
execution,
and
this
one
should
be
child
of
multiple,
multiple
parent,
Trace
traces.
E
B
So
Tyler
is:
is
this
still
a
subject
of
ongoing
discussion
in
the
messaging
seg
I
see
yeah
much
of
the
successions
happening
in
late
November
early
December?
Okay,
so
should
this
group
try
to
come
up
with
our
opinion
on
how
that
should
be
handled
and
communicate
that
to
the
message
instead,
or
should
some
of
the
people
from
this
group
start
participating
in
the
messaging
Sig
to
become
part
of
those
conversations
directly
and
then
report
back.
C
So
I
would
say,
participation
in
the
messaging
Sig,
at
least
for
a
short
period
of
time,
would
probably
be
beneficial
in
the
Sig
meeting
or
in
some
of
the
the
link
discuss
discussions
that
I've
shared
I've
been
attending
both,
but
perhaps
someone
from
the
X-ray
team
that
has
worked
with
this
more
directly
would
make
sense
to
also
attend.
C
I
think
that
we're
hope,
I
I,
hope
that
we'll
get
around
to
discussing
this
particular
topic
this
week,
we've
kind
of
been
postponing
it
a
little
bit,
or
at
least
it
hasn't
been
hasn't
made
it
to
the
top
of
the
agenda
thus
far,
and
we've
run
out
of
time
to
discuss
for
the
past
couple
weeks,
but
this
week,
hopefully
we'll
get
around
to
it.
C
So
I
don't
know
Lubin.
Maybe
you
have
time
to
attend
the
messaging
Sig
meeting
this
week.
If
that
would
make
sense
to
share
your
perspective,
there.
B
Yeah
I
think,
if
we're
looking
for
x-ray
participation,
I
can
reach
out
to
William
armors
and
see
if
someone
from
his
team
is
able
to
participate.
B
That's
the
the
answer
key
and
I
believe
his
team
is
also
handling
the
the
segment
linking
implementation.
That
Lubin
was
describing.
C
Great
yeah
that
would
be
awesome
to
have
someone
from
from
that
team
represent
that
meeting
at
that
meeting.
D
Yeah
I'm,
not
muted,
yeah,
exactly
I
I
just
wanted
to
to
call
this
out
real
quick,
I
I
said
it
before,
so
we
have
in
the
JavaScript
documentation
a
little
bit
on
AWS
and
now
just
recently
added
a
little
bit.
With
the
help
of
my
one
of
my
colleagues,
we
added
a
little
bit
on
gcp
functions
as
well.
D
It's
it's
very
superficial
right.
It's
really
only
like
hey
go.
There
create
a
function,
set
it
up
at
open,
Telemetry
that
that's
what
you
get
but
I
said.
The
intent
was
a
little
bit
to
to
put
it
into
in
into
the
front
of
people,
so
they
can
read
it
and
then
maybe
figure
out
like
hey,
that's
something
I
can
do,
or
maybe
they
run
into
some
issues
and
then
can
decide
to
say
like
hey.
D
Maybe
I
want
to
contribute
back
some
some
improvements
to
that
says
that
I
I
think
we
also
have
to
discussion
if
and
where
the
the
serverless
docs
should
live
on
open
Telemetry.
I
o
I,
don't
think
we
should
have
them
at
a
top
level
right
now
right,
but
but
having
them,
let's
say
a
little
bit
hidden
like
the
the
JavaScript
documentation,
I
think
that's
quite
good
to
to
experiment
a
little
bit.
D
B
D
Exactly
yeah
this,
this
is
more
like
hey
here's,
some
some
native
way
of
doing
it
I
mean
hopefully
at
some
point.
We
can
also
add
some
something
on
the
layers
right,
so
that
that's
definitely
something
that
should
be
there
eventually.
But
I
said
the
initial
intent
was
really
more
like
having
it
there
in
and
calling
out
to
people
like
hey.
You
can
also
do
serverless
with
with
open
telemetry.
D
B
B
No
I'll
just
say:
I
I
think
that's
a
great
start
and
hopefully
we'll
serve
to
draw
on
some
more
people
as
well.
To
have
them
realize
that
it's
it's
a
possibility,
even
yeah.
D
Yeah,
exactly
that's
also
what
I
wanted
to
say,
just
just:
let's:
let's
have
it
there
and
see
how
and
if
people
react
on
it
and
then
grow
from
there.
B
Okay
and
then
the
last
item
I've
added
here,
I
wanted
to
discuss.
B
Actually,
let
me
find
a
link
real
quick,
so
Tyler
has
raised
an
issue
regarding
the
some
language
in
the
spec
that
it
says
basically
that
if
an
x-ray
context
exists
in
the
environment,
it
should
be
used
in
preference
to
any
other
context,
and
so
we've
talked
a
bit
about
the
event
to
carrier
and
how
we
can
kind
of
decouple
some
things
around
how
how
to
identify
what
to
use
as
a
carrier
for
context
extraction,
but
I
think
the
the
question
remains.
B
When
should
we
be
preferring
the
X-ray
context
that
exists
in
the
land
execution
environment
over
an
x-ray
context
that
might
exist
in
an
event
or
a
carrier
and
Lubin
I
was
hoping.
Maybe
you
might
have
some
insight
into
why
that
language
might
have
existed
in
in
this
spec
or
insight
into
how
this
is
intended
to
be
handled
by
x-ray
or
other
things
that
are
using,
that
context
from
the
land,
execution,
environment.
E
Okay,
so
I
know
that
there
is.
There
is
some
work
from
from
on
x-ray
team
to
be
able
to
handle
WTC
context,
propagation
and
four
months,
so
what
I
heard
from
them
that
their
their
ultimate
goal
is
to
make
gratefully
to
w3c
context
using
to
use
it
internally
as
the
definite
a
definitive
of
math,
not
sure
why
this
was
put
in
the
in
the
cloud
version.
B
A
B
The
environments
comes
from
so
is
that
extracted
by
the
land
over
in
time
from
the
event
or
is
it
from
the
event
invocation
API.
E
So
here
is
what
happens
on
the
phone
tank.
The
content
first
tries
to
find
XA,
he
had
to
say:
okay,
this
is
part
of
the
trace
and
if
it
doesn't
find,
it
makes
a
sampling
decision.
Okay,
should
we
sample
this
this
invocation
and
eventually
decide
that
okay
yeah?
E
If
it's
going
to
sample
data,
made
this
database
Lambda
segment
and
create
a
child
segment,
that
is
the
AWS
Lambda
function
and
sends
it
to
the
to
the
to
the
worker
workers
where
the
the
location
is
executed.
E
So
how
the
runtime
works
is
that
when
it
when
it
starts,
it
makes
http
call
like
get
next
and
this
get
next
is
it
is
going
to
get
the
the
next
invocation
and
I
think
this
is
propagated
to
to
go
ahead
to
the
runtime
and
runtime
extracts
it
and
sets
it
as
an
environmentally
able
to
the
bank
to
to
everybody
else
to
use
it.
So
it's
easy
to
use
environment
very
low
in
London
because
there
is
actually
no
concurrency
in
in
the
runtime,
so
it
works
anyway.
B
E
So
it
is
guaranteed
that
it
is
linked
back
to
database
Lambda
segment,
the
the
the
one
that
the
our
front
end
generates.
Then
there
is
in
terms
of
a
current
logic
of
extraction.
There
is
very
almost
nothing
there
like
it
is
just
propagating
and
creating
some
some
new
identifiers.
There.
B
Okay,
so
I
wonder:
if
that's
something,
then
that
should
be
linked
to
whatever
Spain
we
extract
from
the
event.
If
we
do
rather
than
another
option
for
what
could
be
the
parent
span.
B
E
Yeah,
so
just
to
point
a
few
things
here
that
it
the
it
is
not
only
in
Lambda
that
makes
a
tracing
decision
if
somebody
Upstream
had
made
a
casing
decision
and
propagated
context
to
to
Lambda
will
propagate
this
to
the
runtime
so
yeah,
even
if
there
is
no
x-ray
emitted
from
London.
Actually
we
we
don't
have
these
case.
If
we
get
a
segment,
we
always
send
it
to
excite,
but
there
is
also
questions
now
in
that.
E
E
So
in
this
case,
even
if
okay
in
this
case,
what
will
happen
is
that
if
there
is
an
upstream
segment
and
Lambda
was
put
in
this
mob,
Dr
multimet
tracing
it
will
just
propagate
or
whatever
was
coming
and
then
maybe
it
is
worth
to
link
it
somehow
to
in
the
local
television,
somehow
link
it
to
this
x-ray
segments
that
eventually
somebody
else
outside
of
quanta,
okay,
yep.
B
Okay,
so
it's
it,
it
sounds
like
if
the
user
is
using
x-ray
context
propagation,
there's
a
reasonable
chance
that
that
is
going
to
be
the
correct
context
to
use.
But
if
they're
not
using
x-ray
context
for
propagation,
it's
almost
certainly
not
going
to
be
the.
What
they
want
to
use
is
the
parent
of
their
spins,
but
maybe
you
should
be
linked
to
the
display
of
the
instrumentation
creates
for
them,
because
there
may
be
additional
information.
B
Like
you
know,
say
it
came
through
API,
Gateway
and
API
Gateway
had
a
segment
that
that's
attached
to
that
x-ray
segment,
but
but
not
to
their.
You
know,
w3c
or
B3
context,
foreign.
B
What
do
you
think
about
that
about
suggesting
that
the
the
context
from
the
environment
used
to
is
used
to
link
spans
rather
than
to
determine
the
parent.
C
So
I
don't
have
as
much
of
a
problem
with
that
option
because
a
span
can
have
multiple
span
links,
and
so,
if
we
just
always
say
the
environment,
variable
is
added
as
a
span
link
I,
don't
have
a
problem
with
that,
because
it's
not
going
to
you
know
come
as
a.
B
Right
yeah
I
mean
it's
a
couple
hundred
bytes
added
to
the
payload,
but
that's
probably
not
significant.
C
Yeah
I
mean
it
if
people
are
concerned
about
that,
then
maybe
have
a
separate
setting
to
turn
that
off,
but
the
the
logic
to
support
that
should
be
fairly
simple,
as
opposed
to
having
to
prioritize
different
propagation
mechanisms.
C
So
that
that
makes
sense
to
me
having
the
environment
variable
propagate
as
a
span
link.
A
B
It'll
simplify
things
for
what
the
instrumentation
has
to
do
after
it's
called
event
to
carrier
yeah.
It
might
not
need
to
augment
that
with
any
anything
additional
foreign.
B
Okay,
I
I
think
that
helps
give
us
a
path
forward.
There.
C
So
can
we
make
that
as
a
decision
or
do
we
still
need
to
have
more
discussion
around
having
the
environment
be
spendling
instead,
I.
B
I
think
that
we
should
recommend
that
decision
to
the
other
group
next
week
or
asynchronically
like
if
you
want
to
do
this
now,
I
I
think
we
should
recommend
that
as
a
decision
to
be
taken
and
get
some
broader
feedback
on
yes,
no,
we
should
do
this
or.
C
Not
got
it
so
I'm
fine
with
just
having
it
as
a
discussion
point
for
next
week.
C
E
Oh,
so,
let's
continue
this
card
this
next
week,
but
I
wanted
to
point
out
so
that
obviously,
are
not
the
the
best
person
to
represent
x-ray,
but
from
our
discussions
they
were
thinking
something
like
okay.
We
are
now
using
x-ray
context,
but
eventually
we
want
to
start
accepting
w3c
and
B3
headers
and
propagate
them
and
obviously
only
one
context
will
be
propagated
in
Lambda,
but
something
like
okay.
E
There
is
already
a
w3c
context
in
the
invoke
called
Lambda
who
propagate
this
and
eventually
that
will
go
to
the
same
environment,
variable
mechanism,
so
I
think
here
it
is,
and
obviously
now
you
are
creating
the
the
ram
to
the
instrumentation
that
the
function
is
creating
its
own
spouse,
I.
Think
if
the
the
parent,
the
environment,
is
directly
C
segment
and
it
it
will
be
noted
because,
like
the
environment,
actually
it
could
be
noted,
I'm,
not
sure.
Now
it
is
it.
E
We
have
to
double
check,
but
it
is,
you
could
distinguish
it.
Is
it
B3?
Is
it
x-ray
digital
3C?
So
the
the
K
my
case
here
is
that,
okay,
if
there
is
a
w3c
context
in
in
the
in
the
environment,
about
the
instrumentation,
most
probably
we
want
to
put
its
own
spouse
as
a
as
a
children
of
of
that
Upstream.
E
Okay,
if
it
is
an
x-ray,
then
maybe
linking
is,
is
what
is
segment
linking
is
what
what
would
be
best.
B
So
how
about
this
then?
How
about
if
the
instrumentation
were
to
extract
the
context
from
whatever
carrier
the
event
to
carrier
implementation
gives
it
and
also
extract
the
context
from
the
environment
and,
if
they're
the
same.
Obviously
then
just
use
that
as
the
parents
and
if
they're
different,
then.
E
Would
work
so
I'm
missing
a
little
bit
this
context,
propagation
with
carriers?
Etc
is
there
maybe
in
the
notes
here,
maybe
put
a
link
to
to
the
specification
and
what
the
discussion
is
happening
because
I
I'm
a
little
bit
yeah
I,
don't
don't
understand
the
details.
There.
B
B
B
B
Given
the
event
you
know
that
came
in
as
a
you
know,
a
raw
blob,
you
know,
slice
of
fights
or
whatever
it
is
given
that
hands
back
to
me
now
something
that
I
can
use
to
extract
a
context
from
which
is
typically
a
dictionary
or
an
object,
a
map
of
string
to
some
value
and
depending
on
your
language.
But
then
the
context
propagation
will
be
applied
to
that.
B
It
would,
depending
on
what
propagators
are
configured
look
for
w3c
Trace
contact,
Setter
or
for
x-ray
header
or
for
B3
header
at
the
right
keys
in
that
carrier.
E
B
One
correct
yeah,
so
the
assumption
is
that
there
may
be
a
trace
context
embedded
somewhere
in
the
payload
like
it
could
be
a
custom
user-defined
structure
that
looks
nothing
like
an
HTTP
event
or
anything
like
that,
but
still
has
a
context
in
there
somewhere,
and
so
we
give
the
user
a
mechanism
to
extract
that
information
from
their
event
and
put
it
into
something
that
a
context.
Propagator
can
recognize.
E
Yeah
yeah,
that
makes
perfect
sense,
but
then,
if
we
go
back
to
the
logic
that
okay,
if
they
March
use,
use
this
as
a
parent,
if
they
don't
match
up,
it
has
added
as
a
as
a
link
link
segment
the
segment
link
Etc.
There
is
the
third
case
where
there
is
environment.
E
There
is
a
context
in
the
environment,
but
the
the
context
to
event
karaoke
event
didn't
find
any
any
context,
and
in
this
case
there
is
two
options:
either
just
link
it
and
it's
fine.
If
it's
a
x-ray,
maybe
it
is
also
fine
if
it's
a
native
Petri
or
w3c.
But
my
point
is
that
you,
if
you
know
that
okay,
the
environment,
the
the
context
in
the
environment
is
w3c,
you
could
use
it
also
as
a
parent.
B
Yeah
and
I
I
think
what
I
would
suggest
doing
in
in
order
to
extract
context
from
the
environment
is
to
first
use
the
the
properties
that
are
configured
by
the
user,
to
attempt
to
extract
the
context
from
the
environment,
and
so,
if
the
user
is
intending
to
use
w3c,
they
would
configure
w3c
propagator
and,
if
that,
if
that
fails,
then
use
an
x-ray
propagator
as
a
last
resort
to
say
yeah.
We
expect
that
there
will
be
an
X-Ray
context
in
here.
If
nothing
else.
Okay,.
B
I
think
that
gives
us
a
some
good
things
to
proceed
on
from
there
Tyler.
What
do
you
think
we
should
do
as
as
next
steps
should
we
start
with
the
graphic
FTR
to
change
the
existing
spec
language,
or
do
we
want
to
try
to
refine
something
in
the
lender?
Repo
first
regarding
this
and
then
forklift,
that
into
the
spec.
C
B
Okay
and
I
think
with
that
we
have
reached
the
end
of
the
agenda.
Is
there
anything
else
that
anyone
here
wishes
to
discuss.