►
From YouTube: 2022-06-16 meeting
Description
cncf-opentelemetry@cncf.io's Personal Meeting Room
A
A
A
Conference
yeah,
it's
it's
like
feature
stack
basically,.
B
We
already
did
it,
it
was,
it
was
in
va
las
vegas,
as
that
must
be
the
place
to
throw
conferences.
It
is.
B
And
I
think
it
was
it
was
it
was
hybrid,
but
there
was
a
large
in-person
attendance,
so
that
was
kind
of
emphasized
yeah.
A
lot
of
people
were
excited
to
be
back
in
person.
That
sounds
exactly
the
same.
Yeah.
A
There
is
somebody
just
as
a
heads
up
who
is
not
on
our
team
at
splunk,
but
is
a
splunker
who's
interested
in
instrumentation,
who
is
hoping
to
submit
kafka
related
pr,
and
I
know
that
your
kafka
pr
has
been
out
there
for
a
minute.
So
it's
it's
different.
It's
emitting
some
metrics,
I
think
around.
Well,
no
yours
is
metrics.
I
forget
I
forget
which
one
it
is,
but
just
be
aware
that
there
might
be
an
implementation
pr
from
a
splunker.
A
B
No,
it's
so
probably
best
left
to
trust
yeah.
It.
B
E
I
think
he
was
commenting
something
when,
when
it
was
something
like
6
a.m,
important
time,
I
don't
even
know
whether
he's
at
home
currently
that
he's
travelling.
B
Well,
you
mentioned
that
he
might,
he
might
chime
in
on
pr's
and
issues
if
he's,
if
he's
bored
and
feeling
up
to
it.
So
I
guess
we'll
see
if
he's
feeling
like
he
has
enough
steam
to
do
the
release
so.
A
B
I'm
working
on
well,
I
suppose
I
can
just
give
everyone
a
heads
up
so
like
for
the
next.
Between
now
and
the
next
release,
I'm
gonna
be
trying
to
get
exponential
histograms
in
a
better
place.
They're
not
stable
in
the
metrics
specification
yet,
and
I
think,
there's
kind
of
like
a
catch-22
situation
where
they're
waiting
for
prototypes
that
are
so
that
they
can
tooth
two
prototypes
that
people
feel
good
about.
So
they
can
give
it
the
thumbs
up
and
mark
it
as
stable,
and
so
javas
is
okay.
But
it's
it's.
B
It's
really
hard
to
use
in
the
wild
right
now
in
real
applications.
It's
just
really
hard
to
configure,
so
I'm
gonna
be
taking
some
steps
to
get
it
closer
to
being
ready
to
use,
and
you
know
able
to
be
used
in
an
experimental
fashion.
So
the
hope
would
be
that
you
could
configure
a
environment
variable
or
system
property.
That
would
allow
you
to
indicate
that
you
want
to
use
exponential
histograms
when
you're
exporting
metrics
over
otlp,
and
so
there
should
just
be
a
flag
that
you
flip.
B
Where
you,
you
know,
you
switch
from
explicit
bucket
histograms
to
exponential
histograms
for
all
situations,
so.
A
B
So
you
on
the
instrumentation
side,
you
use
the
same
instruments,
but
how
those
are
aggregated
and
exported
is
changed.
B
Exponential
histograms
are
they're,
pretty
wild
and
they're
they're
fantastic,
so
they
automatically
adjust
their
their
scale
to
reflect
the
the
range
of
values
that
have
been
recorded,
and
so
you
know,
we've
received
some
feedback
that
it's
it's
from
folks,
trying
to
use,
histograms
that
you
know
they're
the
values
that
are
being
recorded
aren't
in
the
ranges
of
the
default,
histogram
buckets
and
so
they're,
not
particularly
useful.
B
Sometimes-
and
you
know
you
can
use
views
to
reconfigure
the
buckets
of
explicit
bucket
histograms,
but
you
know
it's
kind
of
it's
kind
of
cumbersome
to
write
a
extension
where
you
programmatically
configure
the
views
so
exponential
histograms
solve
all
those
problems
pretty
elegantly.
E
B
Yeah,
so
I
well
supposedly
well,
I
guess
so.
Otlp
has
a
special,
a
different
data
type
for
exponential
histograms
versus
explicit
bucket
histograms.
B
But
you
know
if
you're
a
back
end
that
says
that
you
are
otop
compliant,
then
you
should
be
able
to
accept
exponential
histograms
because
they're
part
of
the
stable
specification
this
stable,
like
protocol
specification,
the
sdk
specification
is
still
experimental
but
yeah.
So
some
some
back
ends,
I
suppose
won't
support
those
or
be
able
to
support
those
right
away.
E
B
E
Probably
it's
something
that,
like
maybe
they
want
to
add
like
an
attribute
like
what
service
was
actually
called
or.
A
E
I
was
thinking
about
it,
but
the
the
problem
with
spam
processor
is
that
I
think
you
have
to
add
it
inside
the
agent.
E
If
you
can't
do
something
that,
like
that,
the
user
code
writes
something
into
the
context
to
the
spam
key
and
the
processor
is
with
the
key,
because
they
both
are
using
like
the
different
version
of
the
api
like
one
is
using
the
shaded
version
and
the
other
one
is
using
the
version.
That's
bundled
in
the
application.
E
You
would
need
to
breach
the
keys
somehow
like
it
isn't
like
that
easy.
E
References
like
why
I
was
thinking
about
this
is
that,
like
it
definitely
can
be
implemented,
but
the
implemented
this
is
kind
of
a
bit
painful,
especially
considering
that,
with
the
current
extensions,
you
can't
add
anything
to
the
boot
class
back
easily.
E
A
Do
we
yeah
so
there's
I
mean
we'd,
have
to
add
some
kind
of
instrumentation
api
that
users
could
access
right.
We
do
have
that
thing
for
headers
right
where
people
can
specify
headers
that
they
want
extracted
and
put
into
the
into
the
attributes,
but
that's
kind
of
a
single
purpose.
You're
I
mean
the
thing
you're
describing,
I
think,
is
more
general
purpose
and
people,
I
think,
have
this.
This
use
case,
sometimes
of
like
wanting
to
log
an
internal
like
transaction
id
or
some
other
sort
of
identifier
that
relates
to
their
internal
processes.
E
Like
maybe
the
api
could
be
something
that
basically
allows
pushing
a
spam
processor
onto
the
context,
and
if
you
spend
it
like,
if
a
nested
span
is
created,
then
we
just
would
run
that
spam
processor
also,
so
they
could
do
like
whatever
they
want
with
the
attributes.
A
I'm
not
following
that
design
like
can
you
tell
me
more,
like
the
user
we'd
provide
an
api
that
the
user
would
register
their
own
spam
processor
into.
E
Yeah
like
something
that
would
be
called
like
when
the
spanish,
so
they
could
modify
that
to
the.
E
It
doesn't
have
to
be
like
the
span
processor
as
we
have
it
now.
It
could
be
something
that's
that's
similar
to
it
and
inside
the
agent
we
will
have
a
spam
processor
that
looks
into
the
context
whether
whether
it
has
something
there
and
if
it
has,
then
it
just
delegates
the
call.
A
A
Yeah,
exactly
people
want
to
add
dynamic
attributes
to
metrics,
which
some
people
call
dimensions.
B
Well,
that's
sort
of
possible
today,
so
there's
this
experimental
apis
so
that
you
can
indicate
keys
for
a
baggage
which
get
extracted
and
added
as
attributes.
Oh,
I
didn't
know
that
existed.
B
Yeah,
so
that
that
was
something
that
was
punted
on
in
the
initial
1.0.0
of
the
metric
specification,
but
josh
zereth
was
kind
of
pushing
on
that.
So
you
know
it
could
end
up
as
part
of
the
metric
specification
at
some
point.
C
B
A
A
B
Well,
and
if
you
want
so,
if
you
want
a
an
attribute
appended
to
every
single
metric,
that's
exported,
like
I'd,
advise
a
different
approach
for
that
I'd.
Advise
you
wrapping
an
exporter
and
like
mutating
the
metric
data
to
include
the
additional
attribute
before
you
delegate
on
to
whatever
your
base.
Exporter
is
yeah.
A
D
D
B
Yeah,
we'll
give
everyone
some
time
back
so
yeah.