►
From YouTube: 2023-02-02 meeting
Description
cncf-opentelemetry@cncf.io's Personal Meeting Room
A
All
right
did:
does
anybody
here
claim
ownership
of
this
topic.
A
It
was
on
the
agenda
last
week
also
and
nobody
claimed
it
so
we
we
discussed
it
briefly,
but
that's
I
think
that's
fine,
Jack
added
a
couple
things
about
the
123
release,
which
would
go
out
a
week
from
tomorrow.
A
A
Yeah
I
did
see
that
the
the
log
API
is
going
to
be
stable
soon,
which
is
great
at
I.
Don't
know
if
the
SDK,
if
they're
going
to
do
it
in
stages,
like
they've
done
with
the
tracing
API
or
if
they're
going
to
do
them
both
at
the
same
time.
E
C
Yeah,
so
this
is
something
I
remember
brought
it
up
last
night,
but
one
thing
I'm
looking
at
is
to
get
some
Header
information
and
and
I
think
it
was
materials
I
buyed
on
on
slack,
but
I
found
it
before
he.
C
He
got
a
chance
to
respond
to
it,
and
so
you
have
all
the
automatic
instrumentation,
but
you
have
some
extra
headers
or
app
variables
that
you
want
to
get
in
at
as
attributes
and
and
it
seems
like
we
have
sort
of
an
experimental
feature
to
sort
of
provide
a
list
of
headers
and
you
you
can
get
them
added,
but
that
got
me
thinking
a
little
bit
more
and
it
was
something
I
was
thinking
about
with
Spanish,
which
the
first
time
I
saw
them
right
is
if
it
would
be
interesting
to
expand
the
capabilities
of
those
so
right
now
it's
it's
sort
of
you
need
to
use
with
span
it's
going
to
set
up
an
instrumenter,
and
then
it
can
pick
up
the
span
attributes
in
there
sort
of
import,
efficiency
and
whatnot,
but
one
thing
I
think
would
be
pretty
powerful,
is
to
be
able
to
just
annotate
with
span
attributes
and
get
them
added
to.
C
Whichever
is
the
current
span,
for
example,
so
I
tested
out
a
little
bit
and
it's
it's
sort
of
doable
and
sort
of
you
can
expand
a
little
bit
on
the
code.
We
have
it,
like
my
proof
of
concept,
is
sort
of
as
minimal
as
can
be,
but
I
can
run
a
simple,
simple
test
with
it.
C
It
might
be
for
performance
reasons.
We
might
want
to
have
sort
of
a
sort
of
another
sort
of
method
level
annotation,
because
the
way
to
find
The
annotation,
otherwise
it
scan
every
method,
every
parameter
of
those
methods.
If
you
have
it
on
a
method,
you
only
need
to
scan
on
on
a
method
level
right.
C
It
could
do
sort
of
iterate
over
those
parameters
and
and
do
span
set
attributes
with
a
very
similar
functionality
as
the
instrumentation
instrumental
code
does
today,
it's
more
like
getting
feedback
like
this
is
something
that's
been
thought
about
and
then
like.
No,
we
can't
do
this.
It's
going
to
collide
with
other
Auto
instrumentation
or
no.
This
would
be
something
interesting,
I'll
be
willing
to
spend
some
time
and
see
if
I
can
build
up
sort
of
a
more
sort
of
full
proposal
on
or
sort
of
in
POC
for
implementation,
for
it.
D
A
I
mean
one
reason:
maybe
it
hasn't
been
done
yet
is
that
it
doesn't
save
a
whole
lot
over
doing
it
manually.
A
Whereas
the
with
span
definitely
saves
a
ton
of
boilerplate
on
the
method,
yep
and
then
another
advantage
of
the
span
attribute
on
a
with
span
and
I.
Don't
know
if
we
implement
it
this
way,
but
we
could.
A
A
A
So
something
like
you're
thinking
that,
like
with
current
span.
C
Basically,
yeah
foreign,
it's
sort
of
at
least
in
mind.
You
have
some
that
are
sort
of
they're,
more
willing
to
add
an
annotation
because
they're
in
the
spring
world
and
you're
basically
code
by
annotation
right
rather
than
to
to
add
code
in
there,
so
that
would
be
sort
of
the
the
main
reason
for
it.
I
agree
right:
it
doesn't
sort
of
reduce
the
amount
of
things
you
put
in
there,
but
for
from
drama
aspect,
you
keep
your
code
a
little
bit
cleaner
with
sort
of
okay.
C
A
C
C
C
But
then
you
might
have
some
arm
like
you
can
see
the
parameters
on
the
path,
but
we
don't
really
extract
them,
but
in
our
case
we
have
some
headers
that
contain
like
okay,
here's
the
the
account
ID
that
hit
this
and
those
really
don't
show
up
unless
we
explicitly
add
them,
and
it
would
be
nice
to
get
them
in
there
because
then,
in
the
pipeline
later
on,
we
can
create
sort
of
metrics
from
these
sort
of
kind
of
nicely
from
from
response.
C
If
we
wanted
to
with
the
sort
of
the
customer
as
an
attribute
for
that
metric,
for
example,
so
we
can
look
at
a
HTTP
duration
from
sort
of
different
customers
or
different
accounts
as
an
example
and
fully
doable
by
doing
spam.
Current
and
sub
attributes
type
of
manually.
But
I
wonder
if
we
like
it
at
least
my
feeling
right
now
is
I
think
we
would
get
quicker
adoption
if
we
tell
them
like
just
add
his
annotation,
and
it
will
show
up
and
respond
for
you.
A
Sounds
very
reasonable
to
me:
I
like
I
I,
don't
mind
annotations
I'll.
Let
John
is
our
famously
anti-annotation.
E
I,
don't
know
I
mean
I.
Think
I
lost
this
argument.
Like
15
years
ago,
like
we
have
a
programming
language,
let's
program
it,
let's
not
invent
another
programming
language
to
program
our
programming
language
in
I'm
like
write,
some
Java,
it's
not
hard
yeah
but
I,
know
I
lost
this
argument
15
years
ago.
So.
F
Let
me
have
my
my
five
cents
on
this,
so
my
profile,
open
tracing
was
like
this,
so
it's
relied
on
annotations
to
basically
add
attributes
everywhere,
but
that's
very
limiting
because
it's
static
and
if
you
want
to
add
different
attributes,
you
have
you
end
up
creating
a
bunch
of
different
methods,
each
annotated
differently,
and
so
the
programmatic
way
that
we
have
the
example.
Above
So
spine
current
span
and
set
stuff
it's
much
more
flexible,
and
then
we
do
instrumentation.
F
Nothing
prevents
you
from
creating
an
instrumentation
for
the
library
and
that
that
you
have
and
do
that
automatically
for
you,
so
I
actually
prefer
that
the
way
things
are
done
in
open
Telemetry
on
that
trigger.
A
A
Do
you
think
in
micro
profile
it
that
this
would
or
do?
Do
you
expose
the
with
span
annotation
in
microprofile.
D
F
A
Well,
do
you
know
why
it
was
the
only
option
in
micro
profile
before?
Was
that
maybe
just
to
not
expose
API
surface
from
a
backward
compatibility
worries.
F
So
this
this
was
started
by
people
that
liked
a
lot
the
now
Jakarta
e
specs
that
are
heavy
on
annotations
and
I.
Think
that
was
the
philosophy
behind.
A
Well,
given
that
we
already
have
sort
of
annotation
support,
it
seems
reasonable
if
there's
I'm,
trying
to
think
if
there's
a
downside
from
complexity.
C
The
one
thing
I've
been
thinking
of
and
I
think
it
would
be
the
same
for
the
width
span
is
ordering
of
things
when
this
would
be
more
so
than
with
the
with
span
right,
but
with
the
ordering
of
when
or
how
about
body
in
injects
things
right.
When
you
do
the
manipulation,
you
would
want
these
to
happen
after
all,
the
automatic
instrumentation.
In
a
sense,
if
you
do
with
the
width
span
today,
I
guess
it
might
not
matter
because
you're
creating
a
new
like
just
duplicating
the
instrumental
code.
G
C
Not
with
span
but
I
think
here
you
would
need
an
ordering
because
you
need
you
want
it
to
be
the
this
band
that
has
sort
of
you
want
an
existing
span.
So
if
you
do
span
current
before
the
auto
instrumentation
code
right,
so
if
this
gets
underneath
the
first
or
last
right,
depending
on
where
it
lands
in
that
now
sort
of
magically
expanded
method,
you
need
to
be
a
bit
careful
with
that.
C
So
it's
it's
with
span
like
the
so
it
doesn't
matter
with
span.
If
you
do
a
with
span,
you
don't
need
to
do
this
right,
but
if
you
have
the
auto
instrumentation,
it
needs
to
happen
in
such
a
way
that
it
it
lands.
When
this,
that
was
the
instrumentation
code,
has
been
executed
and
created
a
span
in
the
first
place.
C
Yeah
I'll
take
a
look
and
see
it
see
how
how
and
if
I
can
get
it
working
and
it
looks
reasonable
on
a
share
something
and
one
one,
just
quick
question.
So
there's
two
projects
for
The
annotation
injection,
one
which
was
like
1.0
and
one
was
1.16
or
something,
and
that
the
1.16
is
the
the
later
version
and
the
other
one
is
the
deprecated
one.
C
But
so
I
think
it
was
under
instrument,
yeah
expectation
down
here.
A
A
C
C
C
F
A
B
This
might
be
in
the
like
reactor
or
another
reactive
module,
just.
A
Yeah,
so
we've
got
a
bunch
of
tests,
so
Bruno
like
if
you
have
a
case
where
it's
not
working.
Well,
if
you
could
open
an
issue
for
that,
because
there
was
definitely
work
done
to
make
them
work
with
reactor.
A
Where,
like,
if
you
return
reactive
types,
the
with
span
will
wait
to
end
that
span
until
the
the
promise
is
completed
and
then
there's
some
other
things
so
yeah
yeah.
A
That
would
be
great
to
I.
Don't
doubt
that
there
are
gaps
in
its
support,
because
reactive
stuff
is
very
complicated
from
a
monitoring
perspective,
but
they're.
We
we're
not
aware
of
any
gaps
currently
I.
Think
Matthias
is
that
fair
one.
B
H
By
the
way,
if
it
if
it
works,
you
might
also
want
to
check
it
with
some
performance
tests
because
usually
instrumentation
and
reactive,
no,
you
should.
H
Yeah,
so
if
you,
for
example,
instrument
it
on
each,
you
will
see
that
on
your
performance
as
I
can
guarantee
it.
A
Do
we
have
Plum
Json
here
now?
That
would
be
actually
that
would
be
and
we
do
have
a
benchmark
foreign
that
gets
run
nightly.
A
That
actually
could
be
very
interesting
I.
It
runs
with
and
without
the
Java
agent
Benchmark.
What
is
this
called
overhead
benchmark?
I.
B
E
A
Jack
seemed
to
think
he
would
be
able
to
do
the
releases
still
he's
out
on
parental
leave
right
now.
For
those
who
don't
know.
A
D
G
E
So
if
I
remember,
it
wasn't
necessarily
that
we're
considering
other
bases,
but
that
people
might
not
know
like
when
they
just
say
exponential
histogram
it
doesn't
some,
it
could
be
unclear
what
base
was
being
used.
So
I
think
it
was
just
to
clarify
and
then
I
guess.
If
we
do
want
to
add
a
base
17
exponential
histogram.
We
could
add
that
later.
A
H
It
wasn't
there
a
huge
discussion
about
this
with
the
with
the
spec
group,
like
a
base
2
versus
base,
10
versus
HQ,
histogram
and
I
believe
the
decision
for
space
to
eventually.
G
E
A
Yeah,
those
were
some
really
intense,
math
threads
in
the
spec
debating
histogram
exponential
histogram
bases.