►
From YouTube: 2022-06-28 meeting
Description
cncf-opentelemetry meeting-2's Personal Meeting Room
A
Things
that
I've
submitted,
which
clearly
have
low
priority
the
fit's
not
reported
by
a
real
user.
That's
my
theory
at
least.
A
A
So,
first
I
wanted
to
make
sure
that
that
I
understood
the
hyphen
normalization.
A
Thing
so.
A
What
we
do
in
the
instrumentation
config
is
we
store
it
normalized,
but
then
we
also
when
we
retrieve
we
normalize
the
value
that
the
user
gave
us.
Is
that
what
you're
proposing
for
the
sdk
or
were
you
okay,
cool?
I.
A
Okay,
I
thought
I
wasn't
sure
if
you
were
wanting
to
do
something
more
strict,
that
somehow.
C
A
So
if
it,
if
it
did,
live
in
the
open
telemetry
api,
would
we
access
it
via
the
open
telemetry
instance.
B
C
Because
there's
many
patterns
that
we
have
on
the
instrumentation
api,
we
expect
instrumentation
to
use.
So
it's
like
it
has
to
be
opened
on
kpi,
but
definitely
the
usage
would
be
easier
in
the
open,
telemetry
yeah.
I
think,
because
the
static
method
makes
it
pretty
hard
to
use
config
or
make
something
strictly
with
config.
C
A
B
C
A
C
C
If
we
didn't
have
the
instrumentation
spi's
or
I
guess
that's
a
job
asian
spi,
maybe,
but
if
that
wasn't
there
it'd
be
a
bit
simpler.
I
think,
but
since
we
have
these
two,
it's
pretty
hard
to
reason
about,
I
think
like
a
flag,
the
aws
ssd
or
any
of
these
experimental
flags
they,
even
though
we
would
write
the
instrumentation
to
read
them
using
config,
they
could
only
be
overridden
by
the
job
agent
ski,
but
not
the
auto,
configure
spi.
C
A
C
C
C
C
C
B
B
A
Just
to
make
sure
I'm
following
you
saying
the
config,
so
it
would
all
the
config
customization
would
be
done
at
the
auto
configure
spi
and
config
properties,
and
then,
when
the
agent
creates
its
global
config,
it
would
just
grab
that.
A
B
Oh,
there
is
one
potential
problem
where
we
have
we
re
using
the
static
config
internally
with
an
instrument,
for
example,
for
the
supportability
metrics
or
for
the
thread
debug
flag,
and
I
think
there
are
a
couple
of
places
that
will
need
to
access
the
conflict
even
outside
the
java
agent.
C
B
C
That's
for
that's
for
the
spis,
that's
where
it
matters
right
and
so,
for
example,
in
the
aws
desktop
right
now
we
have
the
hotel
sampler
and
then
we
also
have
the
aws
sdk
experimental
options,
true
on
the
same
spi
right
now.
So
if
we
did,
that
split
that
would
have
to
be
split
into
two
spi's,
which
is
fine
a
bit
annoying.
A
And
so,
as
far
as
back
to
the
putting
it
in
the
ap
is
the
open,
telemetry
api?
A
What
was
the?
What
was
the
objection
to
that.
C
C
B
Yeah,
oh,
I
had
some
another
idea
because
one
of
the
reasons
why
we
can't
use
conflict
properties
as
it
is
now
and
the
java
agent
is
that
it's
in
it's
basically
in
the
sdk
package
and
it
has
sdk
dependencies,
the
auto
configures
spi
module,
I
mean
so
maybe
we
could
just
you
know,
flash
out
another
one
small
module
that
just
contains
the
configure.
Nothing
else.
B
C
B
A
Access
to
it-
or
I
mean
we
could
add,
that
to
all
the
builders.
A
Can
they
get
the
config
properties
from
er
yeah?
Where
is
that.
A
A
B
I
mean
we
could,
but
it's
kind
of
no
solution.
C
A
C
Yeah,
it
was
supposed
to
never
have
stood
the
live,
instrumentation
yeah
so,
but
what
did
come
to
mind
also
is
that
one
of
the
first
users
might
be
the
spring
micrometer
thing
with
instrumentary
api
and
so
double
checking
the
config
integration,
because
they
would
probably
want
to
integrate
with
spring
config.
Somehow
I'm
making
sure
whatever
you
come
up
with
also
works
well
with.
C
C
A
A
But
I
mean
I've
certainly
seen
like
I
mean
the
that
pattern
of
you
know,
passing
in
a
config
object
to
things.
A
Azure
sdks
do
that,
basically,
as
a
way
to
abstract
away
the
config
responsibility
and
to
be
able
to
integrate
with
other
config
systems.
A
B
A
That's
not
a
horrible
pattern,
I
mean
for
the
librarian's
instrumentation
instead
of
having
all
those
properties
programmatically
set
on
the
builder
having
a
config
object
that
also
allows
the
programmatic
configuration
or
reading
from
a
property
source
or
other
alternatives.
A
As
far
as
at
least
providing
stable
instrumentation
api
instrument,
yeah
instrument
or
api
and
config,
and
then.
A
C
A
C
A
Yeah
I
mean
I
like
the
I
mean
splitting
it
off.
I
mean
it's
weird,
but
it
solves
a.
A
A
A
And
then
you
know
the
hyphen
stuff
we
can.
A
A
Yeah
and
kind
of
maps
well
to
our
recommendation
to
users
of
library
users
or
using
the
not
using
the
global
anything
global.
A
Let's
go
on
to
oh
so,
while
we're
on
global
things,.
A
A
A
Okay,
any
different.
A
Cool,
maybe
we'll
I'll,
add
a
comment
or
we'll
just
chat
on
thursday's
meeting
with
jack
see
what
his
parts
are.
B
Oh
yeah,
so
the
next
one
is
mine.
I've
noticed
that
we
have
a
pr
open
for
a
month
and
it
adds
a
sort
of
http
rpc
instrumentation.
B
And
when
I
looked
at
it,
I
started
to
wonder
whether
we
need
to
duplicate
http
attributes
here
in
this,
because
it's
it's
kind
of
similar
to
retrofit.
If
I
understood
that
correctly,
like
you
have
interface
and
you
kind
of.
B
Maybe
one
of
the
lower
classes-
not
this
one
but
oh,
I
was
trying
to
look
at
it
first
from
the
end
or
the
second
from
the
end
like
this.
No
no
anyway,
the
usage
of
it
is
that
you
have
an
interface
and
you
just
add
a
method
called
oh
yeah
test,
something
in
request
line,
get
something
and
oh,
it
also
allows
placeholders
like
http
roots
but
client-side.
B
But
anyway,
since
it's
supposed
to
be
like
something
like
rpc,
but
for
http,
do
we
really
want
to
duplicate
all
the
http
instrumentation.
A
I
think,
and
I
just
sent
yeah
I
I
was
looking
at
this,
and
that
was
what
motivated
my
doc
pr,
because
I
noticed
our
span
suppression
strategy.
Docs
are
wrong.
B
And
you
wonder
whether
we
should
even
have
them
on
that
page,
because
that's
that's
kind
of
a
thing
that
would
constitute
an
advanced
option
for
me.
A
A
Spanking
suppression,
because
we
don't
really
have
support
for
or
good
support
for,
nested
client
spans.
B
C
C
B
A
Oh,
they
managed
to
that's
a
good
sign.
Yeah
managed.
C
A
I
don't
know
how
many,
if
it's
something
like
links
where
you
know
it
may
be
some
time
before.
Do
you
know
anurag
if,
like
jaeger,
zipkin
handle
or
care
even
care,
if
there's
nested
like
client
spans.
A
A
B
I
have
one
more
topic:
trotsky
created
an
issue
about
configurable
log
exporter
provider
in
the
sdk
rapper
and
jack
mentioned
that
we
could
add
it
to
the
auto
configure,
not
spi
artifact.
But
we
already
have
a
mention
of
the
log
exporter
stuff
in
the
spi
repository.
B
B
A
Keeps
comes
up
every
two
months
and
I
think
I
finally
got
the
answer
down.
So
let
me
practice
this
and
honorary.
You
can
tell
me
if
they
tell
me
if
I'm
wrong
our
sd
our
policy
on
the
sdk.
Our
policy
is
that,
while
auto
configure
spi
is
stable
you
in
order
to
use
those
methods
you
have
to
pull
in
an
unstable,
artifact
and
therefore
you're,
basically,
that
the
stability
is
not
guaranteed
anymore.