►
From YouTube: 2022-08-11 meeting
Description
cncf-opentelemetry meeting-2's Personal Meeting Room
A
C
A
C
A
A
A
C
I
thought
we
agreed
not
not
to
use
that
yet
because
the
I
don't
think
I
ever
addressed
what
you
and
I
talked
about.
Well,
I.
B
A
C
It
is,
I
thought,
that's
what
was.
I
thought,
that's
what
I
said
to
do,
but
I
don't
think
andrew.
Oh
okay,
okay,
I
could
be
wrong,
but
I
think
that's
where
it
was
left
potentially.
A
C
C
Get
that
cleaned
up
and
get
it
in
just
because
it's
oh
he
force
pushed
so
yeah.
C
All
right,
oh
yeah,
I
would
like
the
thing
to
work.
The
thing
get
done:
anyways
yeah,
because
I
need
that
fixed
up
as
well.
A
C
Well,
I
know
that
we,
the
the
main
thing,
was
that
we
were
asking
for
the
the
new
team
and
then
two
people
got
merged.
I
think.
C
A
A
The
code
I
mean
that's.
C
A
Exactly
oh
yeah,
I
gotta
go
check
who
those
were
and
I
can
ping
people
on
slack.
I
think.
C
The
older
ones
are
the
more
important
ones.
The
optional
attributes
on
94.
C
A
Yeah
I'll
go
through
these
today
yeah.
I
appreciate
it
so
the
one
other
thing
I
had
before
the
metric
stuff-
I
I
just
had
this.
I
thought
of
the
like
ecto
and
phoenix
and
like
looking
at
the
demo
application
open,
telemetry
demo
phoenix
app
in
the
startup.
There's
calls
to
functions
with
no
arguments
that
are
phoenix
setup,
ecto
setup
actually
ecto
might
have
an
argument,
but
the
question
is.
A
B
I
think
I
think
the
thing
is
that
if
I
recall
like
all
those
setup
functions,
do
is
like
telemetry
attaches,
so
there
is
no
actual
like
application
or
process
hierarchy
or
anything
involved
in
it.
All
it's
doing
is
like
running
code
in
in
the
calling
process.
Actually,
so
there
is
no
like
supervision.
C
Or
anything
yeah
it
would
essentially
like
start
and
exit,
and
then
the
other
part
is
we
can't
pass
along
there
is
there?
Is
the
config
debate
a
little
bit
of
like
people
having
config
and
files
of
like
in
their
config
files
of
like
things
that
the
app
should
like
magically
go?
Look
up
a
little
bit
rather
than
being
passed
explicitly
and
with
the
telemetry
stuff
in
particular,
like
you
would
need
to
they're
like
like
the
additional
attributes
is
like
an
example.
C
A
Is
a
library
it
is
not
an
app,
but
so
it
could
have
a
process
there's
nothing
stopping
that.
I
do
find
that
annoying
and
because
I
looked
into
because
I
couldn't
remember.
If,
because
gen
servers
you
can
return,
ignore
it'd
be
nice.
If
applications
you
could
also
return
ignore
instead
of
a
pid,
so
I'm
going
to
bring
that
up
with
the
otp
team,
but
an
empty
process
that
doesn't
do
anything
wouldn't
hurt
yeah.
A
I
just
figured
I'd,
bring
it
up
as
a
potentiality,
but
that
I
I
guess
I
guess
I
wouldn't
push
for
it
until
or
if
ever
otp
team
accepts
something
that
allows
it
a
boot
and
not
start
a
supervision
tree.
C
So
in
our
application,
tree
quote:
there's
just
a
single
dependency
and
you
can
also
like
I've.
Seen
like
you
know,
like
keith
lee
has
like
a
pattern,
usually
of
like
calling
like
a
thing
called
like
telemetry
he's
like
a
file
called
like
telemetry
and
then
does
everything
through
there
and
you
kind
of
see
it
and
like
finish
and
stuff,
but
like
I
don't
think
the
intention
has
ever
been
like.
C
A
It
wouldn't
be
all
the
way
there,
but
it
would
be
basically
automatic
instrumentation.
If
you
have
the
dependency
it
would.
C
Well,
I
mean
nobody's
done
like
a
magic
roll-up
one
yet
either
like,
I
think,
there's
there's
definitely
things
in
the
in
the
spec,
I
think
even
of
you
know
essentially
having
more
kitchen
sink
type
libraries
that
so
I
could
just
drop
in
for
a
basic
phoenix
app,
and
it
has
all
that
stuff
for
them.
B
C
A
A
I
mean
for
people
that
would
want
something
like
this
and
it's
I
don't
know
I
get
the
impression,
especially
at
splunk.
Maybe
it's
different
in
the
elixir
and
I
think
it
is
in
early
community,
but
it's
so
tiny.
Who
cares,
but
the
you
know
elixir.
I
don't
know
if
it
is
in
corporations:
com,
larger
companies,
they
really
like
the
auto
instrumentation
stuff.
I
mean
they're
going
all
the
way
with
a
there's
a
now
an
ebpf
go
automatic
instrumentation
because
you
can't
do
it
and
go.
A
A
People
don't
want
that.
Like
I
mean
they,
don't
want
the
easy
thing
to
drop
in,
they
want
to
make
sure
they've
touched
everything
usually,
but
that
may
be
elixir
because
it's
used
to
bigger
companies
they'd
be
more
interested
in
something
that,
but
but
I
liked
your
idea
of
open,
telemetry,
phoenix,
auto
or
auto
phoenix
that
echo
and
phoenix
and
everything,
because
ecto
is
pretty
much
always
there
and
you
can
technically
check
and
yeah
you'll
get
a
dependency
you
might
not
want.
A
I
don't
know
if
mix
allows
you
to
remove
like
rebar
three,
you
can
remove
dependencies
of
depths,
but
the
yeah-
it
just
pulls
everything
in
and
does
it
for
you,
so
it
doesn't
have
to
be
in
open,
telemetry
phoenix.
I
think
that's
a
good
idea
and
I'll
make
an
issue
about
that,
because
somebody
might
is
that's.
The
other
thing
is
it's.
A
We
don't
have
to
do
it,
but
we
can
a
lot
of
times.
People
seem
to.
C
A
I
feel
like
a
lot
of
times.
People
feel
they
are
limited
to
what
is
already
available
and
so
giving
some
some
guidance
on
what
would
be
acceptable
that
we
would
accept
into
like
contributing
stuff
and
is
a
good
is
a
potentially
good
pattern
would
help
people
to
do
it
because
we're
not.
I
don't
think
any
of
us
are
going
to
have
the
time
or
interest
necessarily
in
writing
it.
So.
C
Yeah,
like
everything
that
we
do
is
in
uplink,
so
it's
like
not,
but
I
think
to
the
the
other
thing
is
just
like:
when
people
are
building
these
things
out,
like
instrumentation
libraries
is
like,
and
the
only
tricky
thing
is
with
the
toolbox
type
feel.
Is
you
still
really
can't
be,
having
necessarily
like
calls
to
the
library
itself
and
you
need
to
like
basically
wrap
everything
that's
in
there
and
and
if
application,
whatever
the
thing
is,
which
isn't
in
the
libraries
themselves.
C
C
A
C
C
So
that's
like
a
secret
anything,
but
so,
like
our
our
phoenix
monitor,
you
know
has
this
as
like
the
guard.
We
still
don't
do
anything
with
phoenix
directly
right,
like
maybe
that
kind
of
stuff,
but
then
like
in
the
init
here
like
it
does
the
the
setup.
That's
too,
like
I
use
phoenix
or
like
it
sets
up
the
hotel
part
as
well
as
all
of
our
normal,
like
metrics
for
for
prometheus
and
then
like
the
the
root
all
that
people
have
to
do
then
like
they
still,
they
just
have
to
add
the
depth.
B
Well,
the
problem
is
that
ecto's
repo
is
part
of
your
application.
It's
not
actually
part
of
ecto
and
their
metrics
are
basically
generated
by
your
code,
not
their
code.
You
know
like
you're
you're
using
you,
do
use
ecto
in
your
module
to
generate
a
bunch
of
code,
but
it's
actually
code.
That's
in
your
namespace
yeah.
C
C
So
if
you
named
like
two
repos
differently,
which
is
a
very
common
pattern
for
read,
replicas
and
whatnot
yeah,
you
have
to
like
change
it,
because
otherwise
it
would
come
in
as,
like,
your
library,
dot,
re,
your
library,
repo
and
then
the
next
one
be
like
your
library,
repli
repo
replica
one
repo
replica
two.
But
you
want
it
to
be
the
same
prefix
essentially,
and
then
we
get
that
part
out
of
the
the
metadata.
C
C
Anyways,
like
really
like,
we
have
an
opinionated
thing
right
here
right
and
then,
like
you,
just
pass
in
your
stuff,
and
maybe
it
will
help
if
we
go
ahead.
Not
that
this
is
the
working
group
meeting,
but
go
ahead
and
move
up
link
over
to
beam
telemetry
and
then
out
of
its
own
org.
And
then,
if
we
want
to
settle
around
like
suggesting
people.
C
C
Well,
I
mean
like
once
we
get
the
metrics.
There
might
be
like
definitions
that
happen
in
there
as
well.
Yeah
yeah
that'd
be
nice,
but
it
would
mostly
be
like
like
pretty
much.
Everything
in
a
blink
is
optional,
like
all
of
these
have
default,
implementations
of
init
and
metric
definitions
and
whatever
so
like
it's
all
optional
to
like
change
anything.
C
C
A
B
Yeah,
I
was
also
thinking
it
could
be
possible
if
you
built
some
some
kind
of
like
an
auto
instrument
or
application
into
the
the
base
thing.
But
then
it
only
gets
turned
on.
If
you
have
a
compile
time
config
that
says,
like
turn
on
auto
instrumenting
and
then
you
could
tell
people
to
do
that
and
then
it
would
do
whatever
magic
we
wanted
to
do.
B
But
I
mean
I
I
don't
know
whether
I
would
use
that.
But
if
people
wanted
it
like,
maybe
that
it's
easier
to
tell
them
to
add
that
one
line
of
config
versus
like
paste
these
four
lines
into
your
application.exe
file.
A
Is
there
so
there's
compile
time
config,
but
there's
not
there's
not
really
a
good
way
to
do
configurable
like
dependencies
is
there
like
if
you
have
well,
if
you
have
phoenix-
and
you
include
something
like
uplink
and
it
would
then
get
you
open,
telemetry
phoenix
is
that
does
mix
have
something
like.
B
That
they
would
have
to
know
about
each
other,
yeah
yeah,
you
can't
like
inject
it.
I
don't
think
you
can
like
inject
a
dependency.
Well,
you
you
can
have
optional
dependencies
right.
I
think
overwatch
telemetry
would
have
an
optional
dependency
on
every
library
that
you
might
have
and
then
it
would
like.
Would
you.
C
B
A
I
wouldn't
want
it
to
you.
I
would
want
it
to
be
something
if
it
was
something
like
the
like
some
auto
instrumentation
library.
That
would
be
able
to
say
I
see
you,
you
know
in
your
depths
when.
B
And
I
think
that's
basically
what
uplink
is
doing
here
is
it's
saying,
like
uplink
understands
phoenix
and
recon
and
socket
reno
and
ecto.
A
Well,
I
mean
I
I
just
it
came
into
my
head
just
because
I
know
mix
like
unlike
rebar.
Three,
it's
a
static.
It's
a
config
file
of
a
data
structure
mix
is
a
actual
module,
so
I
thought
maybe
it
was
something
people
did
but
yeah
I
mean.
A
C
C
A
A
A
Yeah,
no,
it
sounds
more
like
it
would
be.
Some
languages
do
where
there's
like,
or
people
provide
like
a
like,
splunk
or
honeycomb,
or
lightstep
provided
cli
tool
for
certain
languages.
That
says
that
we'll
look
at
your
dependency
tree
and
say:
hey,
either
automatically
add
them
into
the
config
file
or,
it
will
say,
hey.
You
might
want
to
add
these
there's
those
two.
A
A
C
I
think
that
would
be
a
good
path
to
be
honest
like
if
it
was
just
a
mixed
task,
but
it's
more
of
like
a
bootstrap
than
auto
instrumenting.
Oh
yeah,.
A
C
A
A
C
A
Yeah,
it
would
be
something
that,
obviously,
if
the
user
runs,
this
task-
they're,
okay,
with
it
changing
it
and
they
can
undo
it
if
they
aren't
happy
that
it
changes
code
and
it
will
say
it's
going
to
change
your
code
with
that.
But
it's
just
not
yeah.
It's
just
nice
that
that
it
is
usually
always
formatted
by
the
big
sport
matter.
So.
A
A
Well,
yeah:
I
think
we
should
get
some
metrics
try
to
get
through
it.
This
time
I
had
one
question
because
they
related
the
metrics
that
I
saw
because
I
saw
your
telemetry
metrics
in
the
uplink
code.
The
telemetry
metrics
doesn't
have
a
does
it
have.
Can
you
set
that
the
values
are
integers
or
bloats
as
part
of
the
metric
definition?
A
C
Was
going
to
go?
Look
that
up
no,
but
what
we
did
we
did
add
a
while
ago
reporter
options
to.
A
B
B
A
A
I
mean
we
need
to
know,
we
need
to
know
the
type.
If
you
want
integer,
I
mean
the
reason.
C
A
C
A
A
Gauge
is
only
observable
so
gauge
you
can
have,
but
it's
the
callback
one.
So
if
you
wanted
like
to
get
the
the
memory
usage
as
a
gauge,
then
that's
part
of
the
callback
to.
A
A
C
A
Have
a
very
hard
stop?
Okay,
that's
fine!
That's
so!
Instruments
are
created
in
hotel,
meter
and
right
now
they
are
just
they're
just
records
or
maps.
I
don't
remember
and
they
are
passed
to
functions
like
add
and
counter
yeah,
it's
a
record
and
it
has
a
type
in
an
integer
float
and
then
that's
passed
back
to
the
sdk
to
actually
do
the
work
so
in
the
api
there's
just
the
the
meter
that
can
create
your
the
records.
That
are
the
instruments
that
you
pass
to.
A
Whatever
type
of
instrument
you
created
counter
histogram
up
down
counter
observables
we're
not
going
to
talk
about,
because
I
don't
know
how
I'm
doing
those,
yet
it
won't
be
hard,
but
yeah.
B
Just
real
quick
are,
we
is:
is
it
important
that
they're
records,
or
do
we
think,
maybe
we'll
have
an
elixir
wrapper
that
turns
a
map
into
a
record
for
you
or
instruct,
or
something
like
that?
Yeah.
A
A
A
Until
meteor
server
is
where
we
track
readers
and
views.
What.
A
Yeah,
okay,
so
readers
are
how,
since
we
have
to
support
statsd
and
prometheus
style,
which
is
push
pull
and
delta
cumulative
differences,
that's
where
readers
come
in,
so
the
server
which
is
what
we
call
the
provider
is,
will
send,
send
them,
send
the
actually
the
server
will
aggregate
or
the
reader,
and
the
reader
basically
just
provides
the
configuration
of
what
kind
of
stuff
it
expects,
whether
it
expects
a
delta
or
a
cumulative
for
a
counter
stuff
like
that
and
we'll
actually
do
the
and
then
the
reader
will
pass
it
off
to
the
exporter.
A
A
So
the
meter
server
is
where
the
views
are
matched
against
these
instrument
recordings
right
now,
the
measurements
others
I
definitely
go.
Maybe
java
does
the
matching
of
views
to
instruments
when
at
startup
and
right
now
this
is
doing
it
when
a
measurement
comes
up
comes
in
and
that
so
technically.
A
But
since
it's
sending
a
message,
it's
not
slowing
any
user
code
down,
but
it
does
mean
that
the
first
measurement
taken
in
the
server
code-
it's
a
little
slower
because
it
has
to
do
the
matching
and
then
then
it
caches
it
as
a
as
a
match.
So
the
next
time
one
comes
in.
It
just
looks
it
up,
but
that's
in
the.
C
A
Debatable
which
way
to
go,
I
think
how
to
match
instruments
and
views,
and
each
instrument
can
match
to
many
views
so
and
for
each
reader.
There
is
a
copy
of
the
view
because
we
might
have
a
cumulative
and
a
delta
reader,
and
so
we
have
to
keep
track
of
the
full
counter
or
you
have
to
reset
it
to
zero
after
each
collection.
A
C
A
A
A
A
It
loops
through
the
matched
views
which
have
per
reader
and
then
each
one
of
those
is
aggregated
so
for
counter
a
sum
aggregation
of
a
counter,
it's
as
simple
as
update
counter,
if
it's
an
integer
and
the
yeah
there's
the
debate
of
whether
it
should
just
try
to
update
this
table
in
the
user
code.
A
Instead
of
sending
a
message,
the
message
was
because
of
needing
to
match
views
and
not
wanting
to
do
that
in
the
user
code,
because
that
could
take
some
time
and
when
I
found
logger
logger
uses
sends
a
message
for
every
log
message
and
it
can
keep
up.
I
think
that's.
A
We
should
probably
be
able
to
do
pretty
well
and
we
can
split
it
so
that
it's,
whether
we
should
update
an
x
table
in
the
user
code
or
we
should
start
spreading
out
and
have
multiple
processes
to
do
the
actual
metric
updates,
instead
of
just
one.
So
there's
a
couple
options.
There.
B
A
Yeah
exactly
I
mean
it
will
change
a
couple
of
things
like
when
you're
collecting
grabbing
all
the
metrics,
because
you
have
to
go
through
multiple
tables,
but
that's
about
it.
A
Hopefully,
that's
a
bit
of
a
rundown
enough
right
now
because
we're
out
of
time,
but
the
yeah
definitely
lots
of
places
for
performance
of
improvements.
It's
trying
to
figure
out
the
overall
scheme
because
of
these
readers
and
views
and
instruments
how
they
interact
together
and
then
we
can
optimize
from
there.
B
C
A
I
made
it
a
message
passed
to
instrument
because
otherwise
we'd
have
to
have
instrument,
creation
do
work
and
then
you
couldn't
as
easily
get
an
instrument
and
I
think
it's
kind
of
a
pain
if
you
have
to
like
keep
track
of
each
instrument.
A
Gotta
play
with,
but
there's
no
elixir
api.
Yet
so
that's
needed,
and
maybe
you
can
review
that
when
I
write
it
and
have
some
time
to
play
with
with
it,
when
I'm
doing
that.