►
From YouTube: 2022-11-29 meeting
Description
cncf-opentelemetry meeting-2's Personal Meeting Room
A
A
B
So
it
looks
like
I,
don't
know:
are
any
other
shop
providers
I'm
going
to
be
showing.
A
Them
shopping
folk
is
actually
the
demonym
of
choice,
which
is
it's
kind
of
a
very
cutesy
one.
Actually
I
don't
know,
I
would
think
you
would
usually
Sam
makes
it
up.
Sam.
B
Has
said
you
can't
make
it
and
he's
apologizing
up
front
Tito's,
which
I
did
see
that
you
were
doing
some
good
work,
securing
our
our
elusive
sponsorship.
A
You
got
to
keep
the
publicity
game
strong
and
and
yeah
paying
paying
homage
to
the
sponsors.
It
was
hilarious
that
at
kubecon
there
was
a
literal
Tito's
vodka
like
happy
hour
area.
It
wasn't
part
of
the
convention,
I
mean
it
was
just
it's
just
in
the
convention
center
there,
but
I
I
did
like
a
double
and
then
a
triple
take
when
I
saw
it
I
thought
there's
there's
no
way,
there's
no
way.
A
This
is
too
good
but
yeah,
but
it's
become
my
favorite
in
joke,
especially
one
that
I
wasn't
even
around
here
and
I
have
no
idea
how
it
actually
really
got
started
and
it
just
I
love
it.
So
a
lot
of
fun.
Indeed
hello,
fellow
Tito's
aficionados.
How
are
you
doing.
C
D
C
Don't
know
that
I'm
an
Aficionado
as
much
as
I
I
shout
out
to
the
sponsor.
D
A
B
All
right,
speaking
of
cease
and
desist,
why
don't
we
go
ahead
and
get
started
with
the
Tito,
specially
recap
and
yeah
see
where
that
takes
us,
so
I
I
dropped
in
the
slack
yesterday,
this
kind
of
SDK
spec
compliance,
PR
and
Andrew
had
some
excellent
comments
and
also
showed
up
in
person
with
again
some
more
comments.
B
I,
don't
know,
I
do
I
feel
like
this.
There
is
some
fairly
strong
wording
here,
at
least
when
I
first
read
it.
B
That
was
kind
of
my
impression
is
that,
like
a
lot
of,
things
are
going
to
have
to
change
I,
don't
know
that
the
intentions
behind
this
are
quite
as
strong,
at
least
that's
what
I
was
getting
from
the
specsig,
but
at
the
same
time
I
still
worry
a
little
bit
about
the
wording
being
a
little
strong,
because,
ultimately,
that's
what
we're
going
to
get
it's
the
words
out
of
this,
and
so
I
think
you
know,
for
the
most
part,
I
think
our
SDK
and
sdks
around
are
doing
a
really
good
job
at
conforming
to
the
spec,
as
in
we
meet
the
requirements
of
this
bank.
B
I
think
it's
where
we
go
above
and
beyond
that.
This
becomes
like
a
little
bit
of
a
concern,
but
like.
B
B
But
I
I
do
I
think
the
the
the
goal
behind
all
this
is
ultimately
I
think
you
know
long
term
for
the
project
when
everything
is
specked,
the
project
would
like
to
kind
of
see
like
some
uniformity
across
the
different
sdks
and
so
that
a
user,
so
users
don't
have
to
like
learn
a
new
way
of
doing
everything
in
each
language.
I
think
it's
ultimately
like
the
goal.
It's
like
the
configuration
should
be
pretty
standard.
B
The
the
Telemetry
should
be
pretty
standard
for,
for
various
libraries,
just
like
things
should
be
pretty
uniform
across
the
board.
There
should
be
like
a
lot
of
surprises
and
you
shouldn't
have
to
like
learn
a
whole
new
paradigm
to
use
languages
from
from
different
sdks
I.
Think
that's
like
the
over
arching
goal.
A
That
was
my
read
as
well.
I
think
that
the
intent
was
to
encourage
more
uniformity
and
not
to
discourage
experimentation,
but,
like
you
mentioned
as
written
it,
it
could
be
very
difficult
to
continue
to
have
any
experimentation,
I
think
you're
right.
The
little
bullet
point
is
intended
to
be
the
loophole.
A
It
would
grandfather
in
the
changes
we've
made,
which
was
nice
because
otherwise
we'd
have
to
go
and
rewrite
a
huge
portion
of
code,
and
that
would
be
very
frustrating
but
yeah
I
think
I
think
it's
it's
a
dangerous
change
and
and
I'm
a
little
worried
about
it.
They
are
going
to
go
back
kind
of
the
drawing
board
and
seek
more
input
and
revise
it
and
try
to
get
somewhere
better
with
it.
A
Yeah
I
think
that
one
interesting
thing
that
came
up
in
the
specsig
meeting
was
that
I
forget
who
was
reviewing
different
metrics
SDK
compliance,
but
he
said
that
only
python
was
actually
compliant
and
then
everyone
else
had
actually
done
something
slightly
different
in
the
spec
and
that
that
was
very
interesting
and
you
can
read
that
as
one
of
two
ways
I
think
you
can
look
at
that
and
say
well
the
fact
that
the
spec
allowed
undefined
behavior
in
a
way
encouraged
people
to
find
the
right
choice.
A
But
the
fact
that
it
allowed
non-spec
compliant
Behavior
also
means
that
those
changes
didn't
bubble
back
up
into
the
spec
discussion
and
didn't
become
formalized
as
the
right
way
to
do
it.
The
spec
didn't
evolve.
So
there
is
a
really
tricky
balance
to
strike
between
allowing
that
experimentation
and
finding
the
right
approach
and
then
making
sure
that
when
people
find
these
things,
we
we
actually
bring
them
back
and
change
the
spec.
So
the
spec
doesn't
diverge
from
reality.
A
So
I
think
that's
a
very
hard
balance
to
strike
I'm
interested
to
see
where
the
pr
goes.
A
The
opinion
that
I
voiced
at
the
meeting
was
I
think
what
would
help
our
more
strict
guidelines
around
how
to
do
experimental
things
and
how
to
mark
them
as
not
part
of
the
official
SDK,
but
something
that
the
you
know
a
given
language
is,
is
working
with
or
trying
out,
I
think
that
could
go
a
long
way
towards.
It.
A
Also
I
think
that
frankly,
I
think
we
could
probably
accomplish
the
goals
of
this
by
using
the
spec
as
it
is,
and
adding
bits
and
pieces
to
say
in
each
specific
section
that
we
care
about
you
know
conformant
apis
must
not
add
any
additional
metrics
view
related
functionality
or
something
like
that.
There
could
be
more
targeted
ways
to
apply
this
I
think
as
well,
so
I'm
I'm
gonna,
follow
that
one
pretty
closely.
A
If
we,
if
we
accidentally
over
correct
and
make
it
too
hard
to
experiment,
then
the
project
could,
you
know,
potentially
face
some
stagnation,
as
people
want
to
go
forward,
but
don't
want
to
be
bound
down
in
process.
So
it's
an
interesting
one
to
keep
your
eye
on
and
if
you
have
more
comments,
people
should
definitely
chime
in
you
know
we
all
work
on
this
stuff
and
everyone
seems
pretty
friendly.
It
seems
like
they'd,
be
open
to
that
feedback.
If
we
haven't
yeah,
there
was
a
big
concern
about
stability
and
backwards.
A
B
Think
definitely
some
excellent
points.
There
I
think
the
the
one
concern
that
I
would
kind
of
echo
I
guess
is
that
it
does
sound
like
if
you
want
to
add
something
that
isn't
part
of
the
spec,
then
the
only
way
to
do
it
is
to
go
through
the
spec
process
and
I
think
we
run
into
this
outfit
where,
like
there
is
some
things
that
that
we
need
to
do
today.
B
That
is
not
in
the
spec
and
we
kind
of
try
to
come
up
with,
like
the
best
compromise,
I
think
usually,
but
even
if
you
are
ready
and
willing
to
do
all
the
spec
work,
it
can
be
quite
a
quite
an
undertaking,
I
think
to
socialize
your
your
ideas
to
the
specsig
and
just
kind
of
get
enough
people
interested
enough
people
to
actually
read
your
stuff
and
sign
off
on
it.
B
So
I
think
like
yeah,
definitely
concerns
about
that
whole
process
and
how
how
big
of
a
challenge
that
is
today
and
how
that
might
impact
things
going
forward.
E
Yeah
I'd
be
curious
to
understand.
Sorry,
oh
my
God
soup,
sorry
recorded
whoever
reads
that
uses
in
six
months.
Yeah
like
we
could
take
an
example
of
like
something
that
made
it
into
the
spec
from
like
three
months
ago
or
something
that's
like
a
guinea
pig
and
be
like
okay,
like
how
did
this
get
into
the
spec
like
what
were
what
were
the
mechanics
and
it
feels
like
the
mechanics
are
usually
like.
E
We,
we
we
rolled
it
and
go
in
a
non-spec
way
and
validated
that
it
worked
against
some
people
actually
using
this
stuff
in
production
and
then
like
brought
that
as
evidence
to
be
like
you
see,
look
this
here's
our
POC
and
it's
live
and,
like
you
know,
GitHub
or
Shopify
is
using.
It
has
used
it
for
six
months
without
issue
and
like
I'm,
not
sure
like,
if
you
retroactively,
attempted
to
apply
this
process
to
the
majority
of
things
in
the
specification
you
know
whatever
you
want
that
to
be
like
exponential
histograms.
E
That
jmxd
has
been
working
on
is
like
maybe
another
example
where
it's
like
you
know
he
shoved
into
the
hotel
collector
a
light
step.
You
know
like
implementation
of
this
and
like
with
the
idea
being
like
it
would
get
replaced
after
it's
proven
out
like
it
just
feels
like
a
lot.
It
just
feels
like
this
will
end
up,
causing
stagnation
in
practice
and
the
motivation
for
95
of
users,
of
open
telemetry
to
get
Cycles
to
push
things
through
the
OT.
E
Providers
and
and
vendors
are
the
only
people
who
will
reasonably
be
able
to
get
stuff
through
the
spec,
because
they'll
be
the
only
people
so
like
I
I
think
this
is
a
difficult
conversation.
I
also
think
it's
important,
because
we've
had
this
disagreement
internally,
where
it's
like
you
know,
I
want
to
YOLO
merge
something
in
a
contrib
and
then
Francis
or
Robert's
like
whoa.
Wait
a
second
there's
a
specification.
You
can't
just
do
that
and
I'm
like
well,
then.
What
are
we
supposed
to
do
and
that's
sort
of
like
up
in
the
air
of
like?
E
Should
you
be
telling
people
to
go
roll
their
own
distros?
Should
there
be
an
experimental
folder
which
gets
stuff
like
a
trace
response,
like
so
I
think
it
would
suck
to
see
this
get
merged
without
clarification
around
the
full
with
around
the
other
stuff,
because
it
all
feels
like
hand
in
hand
and
what,
if
this,
if
this
this
alone
gets
merged?
It's
just
like.
What's
going
to
happen
in
practice,
it's
just
nothing!
Well
we'll
stagnate.
Anyway,.
A
Think
I
am
one
of
the
only
people
I
know
that
randomly
browses
GitHub
every
now
and
again
for
tags
that
I'm
interested
in
to
see
what's
happening
like
people
mostly
don't
do
that
they're
not
going
to
find
it
unless
it's
easy
to
find
and
My
worry,
is
that
like,
if
it's
merged,
as
is
then
the
even
you
know
like
does
having
things
in
the
official
Ruby
repo,
but
in
an
experimental
folder?
Is
that
now,
okay,
or
is
that
not
okay
like
where
there's
it
introduces
more
uncertainty
into
like
that
part
of
the
process
that
I?
E
A
I
think,
like
I
I,
think
that
our
usage
would
be
okay
under
this
honestly,
but
I
also
think
that
I
would
then
have
to
make
that
case,
whereas
before
I,
don't
think
I
needed
to
make
that
case.
That's
kind
of
the
you
know
now,
I'm,
just
not
sure
and
I
don't
want
to
have
the
argument
with
people
over
and
over
again,
certainly
yeah.
E
Yeah
not
to
like
the
answer
here
is
actually
comment:
Upstream,
as
always
well,
I'll
just
ran
a
little
and
just
say
like
one
of
the
big
efforts
and
like
the
big
fud
areas
around
open
Telemetry.
E
If
we
looked
back
a
year
ago,
was
like
oh
things
are
marked
they're,
not
GA
like
there's
this
big
push
for
GA,
so
it
would
feel
like
we're
snatching
defeat
from
the
jaws
of
Victory
a
little
to
be
like,
and
our
solution
is
guess
what
technically
this
stuff
is
GA,
but
in
practice
go
to
this
experimental
folder
for
the
stuff.
You
really
want
like
feels
like
it'll.
Just
we've
just
come
full
circle,
and
then
people
will
be
like
well
wait
a
second.
This
features
experiment
like
can
I
use
it.
E
Is
it
safe
and,
like
so
I,
don't
think,
there's
easy
answers,
but
yeah
I
think
I'm,
just
like
I
just
like
an
answer
so
that
I
don't
have
to
like
make
judgment,
calls
or
we
don't
have
to
make
judgment
calls
and
whatever
it
is.
It
is
but
yeah
having
an
answer
for
the
full
range
of
these
things.
Instead
of
just
like
this,
one
narrow
part
would
be
more
helpful,
all
right.
Okay,
all
right
I
could
have
used
that
time
to
cover
Upstream.
B
Cool
no
I
I
think
this
is
all
really
good
discussion
and
yeah.
Very
good
points
definitely
comment
on
the
issue
and
we'll
kind
of
follow
it
as
it
evolves.
I,
guess
and
yeah
hope
that
what
we
get
is
good
definition
around
how
to
handle
these
situations
without
you
know
overburdened
or
burdensome
kind
of
restrictions.
I
guess
that
would
be
kind
of
the
goal
of
of
the
continued
discussion.
B
B
Against
the
Tito's
time
limit
I
mean
luckily
Sam,
isn't
here
so
we
you
know
we
can
go
a
little
bit
longer,
but
I
will
try
to
wrap
this
up.
So
we
can
get
on
to
our
repo.
The
thing
I
did
want
to
at
least
take
a
moment
to
talk
about
is
Ted,
has
a
process
and
timeline
for
stabilizing
semantic
conventions.
So
this
is
another
kind
of
big.
B
The
big
black
hole
of
of
issues
in
in
hotel
is
that
none
of
our
instrumentation
is
actually
stable
and
someday
users
are
going
to
care
about
that,
and
so
Ted
put
together
this
process
and
I
guess
he
did
some
back
in
the
neck
and
math
at
like.
If
we
just
continue
at
the
current
rate,
he
was
saying
it
will
take
nine
years
to
have
semantic
conventions
defined
for
the
different
areas,
but
he
has
put.
B
This
plan,
basically
a
timeline
there
will
be
certain
focuses
and
they
kind
of
span
a
quarter
and
the
goal
would
be
to
have
signed
a
conventions
to
find
at
the
end
of
each
one
of
these
quarters,
I
think
and
then
at
some
point
when
they
are
defined,
the
language
six
would
carry
out
the
the
implementation
for
them
and,
if
all
goes
well,
it
seems
like
July.
B
1St
2024
would
be
that
the
potential
date
when
this
could
all
be
done
so
I'm
not
sure
how
that
works
out,
because
this
timeline
goes
into
Q3
of
2023.
So
maybe
this
is
just
like
padding
with
an
extra
three
cues
on
the
end
or
for
everything
to
finish
up,
but
this
is
definitely
worth
looking
at
I
think
there
was
yeah
I.
Think
people
were
generally
fine
with
this
idea.
I
think
there
were
some
discussions
about
like
well.
B
How
are
you
deciding
on
which
ones
to
focus
on
during
each
quarter
kind
of
thing
like,
and
there
was
more
of
a
call
for
it
to
be
less
of
like
a
top-down
decision
and
more
of
if
groups
are
ready
to
like
kind
of
work
on
things.
B
For
a
quarter,
why
can't
they
submit
like
a
charter,
saying,
hey,
we're,
ready
and
then
had
the
TC
just
kind
of
approve,
okay,
we're
going
to
if
you're
going
to
do
the
work,
we're
willing
to
do
the
reviews
and
everything
else,
blah
blah
so
I
think
there's
some
discussions.
That
will
probably
happen
about
about
how
this
work.
B
And
we
are
left
with
a
lot
of
questions
like
this
is
something
custom
that
we're
adding?
Is
it
kind
of
Ruby
specific?
Is
there
you
know
a
way
of
addition
uniformly
be
done
across?
You
know,
instrumentations
in
other
languages
and
I.
Think
right
now,
like
all
sigs,
are
kind
of
doing
the
same
thing
we
all
end
up
with
just
like
some
tangled
web
of
configuration
options
to
collect
this.
E
Two
comments:
one
is
I
recall.
There
was
like
at
some
point
some
effort
to
be
like,
let's,
like
oh,
like,
let's
go,
find
Simon
Williamson
to
get
his
opinion
on
Django
semantic
conventions
or
dhh
to
get
his
opinion
on
like
we
were
bringing
in
like
experts
on
their
top
Frameworks
and
things.
Those
are
just
the
two
names
I'm
sure.
There's
many.
You
know
I
lean
to
get
feedback
on
Trilogy
or
whatever,
like
so
is
that
still
a
thing
that
people
are
doing?
It
feels
like
no
I'm
curious
about
that.
E
The
other
was
like
I,
don't
know,
maybe
Francis
or
Robert
we'll
discuss
it,
not
gonna
be
saying
like.
Maybe
some
of
these
are
too
onerous
to
implement
some
of
these
semantic
conventions,
and
it
might
be
nice
to
have
some
sort
of
like
understanding
or
yeah,
never
mind.
I
won't
talk
about
that
second
thing.
Yeah.
Are
we
doing
that
still?
Are
we
like
trying
to
reach
out
to
language
experts
to
like
get
their
framework
experts,
or
was
that
kind
of
died
in
darkness.
B
It
looks
like
here
it
says,
generally
speaking,
the
open,
Telemetry
spec
segment
does
not
contain
experts
in
every
single
domain.
Therefore,
we
will
need
to
find
subject
matter.
Experts
from
The
Wider
Community
to
join
these
groups,
so
I
think.
Definitely
there
there's
a
desire
to
have
some
expertise
in
some
of
these
areas.
B
I
would
say
if
I
were
if
I
were
building
my
semantic
convention,
Dream
Team
I
would
I
would
like
to
have
some
subject
matter.
Experts
people
willing
to
kind
of
navigate
the
the
hotel
side
of
the
spec
work,
but
it
would
be
really
awesome
if
you
did
have
somebody
like
a
first
party
that
provides
you
know
something
from
these
domains
willing
to
be
like
hey,
I
I.
Will
guinea
pig
a
a
version
of
my
library
with
Hotel
baked
in
because
I
feel
like?
That's.
B
That's
really
like
the
you
know,
the
long-term
goal,
and
if
we
can't
get
semantic
conventions
to
a
state
where,
where
a
you
know,
Library
authors
know
how
to
use
it
and
can
use
it,
you
just
need
you
need
to
get
these
things
to
a
state,
I
guess
where
people
will
be
able
to
use
them
and
willing
to
use
them
and
bringing
them
into
the
fold.
B
F
Well,
I
think
it
goes
hand
in
hand
with
whether
or
not
like
we
work
with
people
closely
for
first
party
instrumentation
on
a
lot
of
these
things,
because
it's
like,
if
a
reason,
a
way
to
clarify
what
those
some
com
values
are,
is
to
talk
to
whomever
right.
F
So
like
I
remember,
there
was
like
some
conversation
where
there
was
some
discussion
around
what
GC
semicom
attributes
should
be,
and
it's
like
whether
they
should
be
generic
or
whether
they
should
be
specific
to
the
platform.
There
was
a
lot
of
debate
that
may
or
may
not
have
been
valuable
as
a
result
of
that
right
where
it's
like
the.net
GC
attributes
are
going
to
be
different
than
the
five
different
versions
of
the
jvm
GC
that
are
available,
and
it's
like.
F
Why
would
you
need
to
have
generic
semantic
conventions
for
every
single
one
of
these
things,
even
though
they
implement
the
model
a
little
bit
differently?
So
I
think,
like
you
know,
there's
there's
some
level
of
like
bike
shedding
which
I
think
scares
people
to
like.
F
You
know
get
involved
in
like
trying
to
Define
generic
ones
and
then,
like
figuring
out
ways
for
us
to
work
with
first
party
instrumentation,
to
try
to
come
up
with
reasonable
semantic
conventions
for
those
specific
libraries
you
talk
about
that
with
you
know,
the
examples
that
Eric
was
talking
about
was
like
hey.
We
should
talk
to
rails
committers,
where
else
maintainers
to
figure
out
like
hey
what
would
be
meaningful
semantic
inventions
for
rails
if
we
were
doing
stuff
and
and
then
for
somebody
to
kind
of
like
Shepherd
or
want
to
try
to
bike
shed.
F
A
F
Guess
it
Loosely
follows
model
2,
MVC
right
and
you
can
use
that
for
Django
or
for
you
know.net
rest
or
whatever
it
is
so.
It's
like
it
just
seems
like
just
like
a
lot
of
work,
I,
guess
and
and
trying
to
find
the
people
who
would
be
interested
in
that
kind
of
work.
It's
like
I
think
we
we
might
get
further
if
we
can
get
more
people
involved
in
first
party
instrumentation
of
these
libraries,
because
that'll
trigger
conversations
around
some
comp.
F
B
B
Yeah
we
are,
we
are
running
far
far
over
the
the
allocated
time.
I
mean
good.
B
As
they
have
had
the
sponsorship,
so
any
last
questions
or
any
less
comments
about
semantic
conventions,
if
not
I
will
briefly
briefly
just
mention
this
last
issue
because
we
talked
about
it
a
lot,
but
there
is
a
PR
right
now
with
many
many
green
checks
on
it
that
changes
the
status
of
Json
protov
off
encoding
too
stable.
B
So
so
it's
exciting
I
think
there
are
like
a
few
last
minute.
Clarifications
I,
don't
think!
There's
anything
crazy
going
on,
but
this
is
looking
like
otopjson
will
be
stable
shortly
and
then
I
say
I'll
point
out,
there's
just
a
few
metrics
things
that
people
are
talking
about
for
better
compatibility
or
uniformity
with
open,
metrics,
Prometheus
blah
blah.
B
Yeah,
so,
unless
there's
any
further
discussion,
we
can
maybe
call
the
specsake
recap
done.
You're
welcome,
Tito
the
extra
time
free
at
us
this
week.
We
we
won't.
D
B
B
The
thing
that
is
probably
next
up
on
on
the
agenda
so
Arielle
would
you
like
to.
F
Well,
this
has
come
up
a
couple
of
times,
I
think
where
it's
like,
where
folks
have
or
we've
added
attributes
that
basically
say
like
optionally,
create
Roots
fans
whenever
you're
when
you're
working
with
an
instrumentation-
and
you
know
you
can
essentially
disable
it
and
it
looks
like
you
know-
we
have
another
situation
where
it's
kind
of
like
we
have
these
floating
instrumentation
spans
that
occur
that
aren't
connected
to
a
Trace
by
any
other
means
right.
So
this
isn't
a
situation
where
spans
are
getting
dropped
right.
F
This
is
a
situation
where
you
have
something
that's
in
a
program
that
is
making
calls
out
to
postgres
in
this
example
or
like
something
that
is
pulling
redis
and
it's
like
hey
I'm,
pulling
redis,
but
there's
no
instrumentation
around
that.
So
let
you
know
there
isn't
say
like.
C
F
F
That
would
be
the
the
starting
point
right,
so
you
wouldn't
have
a
consumer
that
is
the
starting
point
of
the
span
or
a
server
span.
That
is
the
starting
point
in
the
span.
That
would
start
triggering
all
of
these
automatic
instrumentation
clients,
bands
to
start
firing
off.
F
And
so
little
by
little
we've
been,
it
looks
like
you
know:
we've
been
adding
whether
or
not
we
should
omit
those
or
we
should
suppress
those
and
I.
Think
it's
because
you
know
like
depending
on
where
you
are
right.
If
you've
got
like
there's
a
couple
things
that
come
into
play
here,
one
of
those
is
sampling
right.
So
if
you
have
a
set
always
on
sampler
right,
there's
going
to
be
some
instrumentations
that
are
like
yeah
just
record
this.
F
Whether
or
not
this
came
from
service
band
or
not
or
for
you
know,
was
started
by
a
consumer
spanner
then
there's
you
know
situations
where
you
are
on
the
client,
but
you
did
not
explicitly
set
in
Span
yourself.
F
You
did
not
create
your
own
Span
in
your
say,
like
you're
running
a
script
or
something
like
that,
whether
or
not
you
added
inspan
does
that
dictate
whether
or
not
a
a
automatic
instrumentation
should
record
something
if
that
makes
sense,
and
so
I
feel
conflicted
about
sort
of
like
adding
continuing
to
add
individual
attributes
to
instrumentation
to
say
whether
or
not
they
should
be
verbose
or
not.
F
F
C
So
this
PR
just
to
to
focus
on
this
PR.
For
a
moment,
this
option
would
flag
the
postgres
instrumentation
that,
if
postgres
is
in,
is
starting
a
trace
like
I'm,
about
to
create
a
span,
and
it
doesn't
have
a
parent,
so
I'm
starting
a
trace.
This
flag
would
say:
no
don't
do
that.
F
C
F
F
A
So
I
he
he
gave
some
rationale
in
in
a
comment
on
the
pr
a
little
bit
further
down
in
in
the
in
the
main
page,
and
the
specific
case
here
is
he's
running
something
in
the
rails
console
and
it's
now
generating
a
span
when
he
loads
like
user.first,
and
so
that's
interesting,
I.
Don't
remember
the
exact
discussion
around.
Why
why
we
did
it
for
redis,
but
at
least
with
redis.
There
are
things
like
sidekick
that
are
pulling
constantly
and
it
creates
a
lot
of
noise.
A
I
think
that,
for
starters,
that's
probably
less
common
for
postgres
in
this
specific
case,
but
not
impossible,
I
mean
there
are
postgres
based
job
cues,
Etc
I.
Think,
like
I,
don't
feel
great
about
adding
this
option
and
I
will
go
say
so
on
the
pr
now
that
I've
seen
it
mainly
because
there's
a
lot
of
other
ways
to
accomplish
this
I
would
really
only
want
to
add
it
if
we
couldn't
come
up
with
another
way
to
suppress
this
there.
A
You
know,
for
instance,
if
you
are
loading
a
rails
console,
there
are
ways
to
detect
that
that's
happening,
and
you
could
not
load
your
instrumentation.
In
that
case,
one
could
argue
that
you,
maybe
you
shouldn't,
have
traces
generated
when
you're
mucking
around
in
the
rails
console
most
of
the
time
which
just
ways
to
detect
that
go
ahead.
C
Well-
and
that
dovetails
to
my
experience
in
trying
to
instrument
Mastodon
for
the
masses
as
a
use
case
of
I,
don't
want
to
generate
traces
from
rails
console.
Maybe
I
want
to
generate
traces
from
rake
tasks,
but
I
know
I
want
to
configure
things
maybe
differently
when
the
web
process
is
firing
up
versus
when
the
sidekick
processes
are
firing
up,
maybe
they
have
different
resource
attributes
applied
to
them,
so
it
could
be
that
maybe
we
have
to
make
it
friendlier
for
folks
like
this,
we
go
and
document.
C
Here's
like
the
generic
rails
initializer,
but
know
that
if
you
just
put
this
SDK
configure
Block
in
your
rails,
initializer
anytime,
you
fire
up
your
rails
app
for,
for
your
actual
app
for
your
rails,
console
anything
that
loads,
the
rails
environment
is
going
to
get
instrumented.
Don't
want
that.
Here's
some
patterns
for
like
yeah.
F
But
to
Eric's
point
there:
this
is
like
a
sampler
shoved
into
a
config
option.
Should
the
person
set
Hotel
traces
to
off
in
those
cases,
but
then
they're
saying
also
that
they
want
to
be
able
to
use
in
Span.
So
should
our
default
mode
be
automatic
instrumentation
only
records
things
if
there's
an
active
span
recording
at
that
particular
moment.
A
C
F
Then
well,
I'm,
sorry
I.
Take
that
back.
It's
like
it's
either
because
because
because
you
know
not
for
Server
spans
for
any
any
span,
that
is
a
client
span
or
an
internal
span
you'd
have
to
that.
That
would
be
the
cases.
A
C
D
C
This
also
raises
I
have
found
in
trying
to
use
our
Auto
instrumentation
speaking
of
documentation.
That's
needed.
Each
of
the
options
in
in
here
need
like.
C
C
E
Yeah
as
an
aside
I,
do
agree,
Rob
that
having
a
gooder
Docs
or
what
the
config
options
are
is
helpful,
and
this
would
be
super
especially
like
a
little
in
my
mind,
I'd
love
a
little
screenshot
of
what
the
trace
looks
like
before
you
turn
it
on
and
what
the
trace
looks
like
after
you
turn
it
on
and
I,
don't
really
care
where
you're
viewing
the
Tracer,
but
you
know
I,
think
I,
think
there's
a
bunch
of
ways
you
can
make
it
better.
E
So
that's
cool
I
think
we're
drawing
a
little
bit
of
a
false
Ariel
alluded
to
this
earlier,
or
maybe
someone
else
most
of
the.
As
far
as
I
know
the
other
places
where
we
have
this,
which
is
like
sidekick
and
Radice,
maybe
I'm
sure
a
few
others,
that's
like
something
in
the
library,
that's
noisy
and
we're
just
you
know,
like
the
library
itself
is
creating
that
noisiness.
This
is
like
a
general.
This
is
just
saying,
like
Samplers,
are
scary,
I,
don't
know
how
to
use
them.
So
therefore,
let's
shove
it
into
conflict.
E
Well,
it
that's
our
fault,
like
you
know,
because
we
haven't
explained
how
Samplers
are
used,
but
this
is
a
general
like
I
feel
like
he's
trying
or
they
are
trying
to
take
a
general
Behavior.
They
want
to
turn
on
and
like
it
just
postgres
just
happen
like
why
postgres
it's
got
nothing
to
do
with
couldn't.
C
E
Like
that's,
not
a
reasonable,
like
that,
isn't
a
reasonable
thing
to
add
as
config
like
it's
just,
not
something
we
should
be
I,
don't
yeah
I,
don't
know
I'm
I'm,
usually
Pro
Plus,
One,
More
config
option.
This
just
feels
like
it
won't
do.
We
should
link
to
a
code
sample
of
how
you
would
use
a
sampler
to
achieve
this.
If
we're
going
to
reject
the
PR,
but,
like
I,
think
this
pair
should
be
rejected.
Yeah.
It's.
C
Either
a
sampler
or
how
to
separate
your
conf,
your
SDK
configuration
and
initialization
in
the
different
ways
that
you
would
fire
the
stuff
up
in
Ruby
like
so.
D
C
A
A
Many
batteries
of
different
sizes
and
you
can
stack
them
all
together,
fun
fact:
several
types
of
batteries
are
actually
like:
coin
cells,
stacked
on
top
of
each
other
and
then
wrapped
up
in
a
wrapper
some
of
the
weirder
sizes,
but
that's
not
neither
here
nor
there.
So
I
had
two
points.
The
the
first
was
that
I
do
think.
A
lot
of
documentation
could
help
this
because
there
are
definitely
ways
to
do
it
without
sampling
and
with
like
careful
initialization
in
rails.
A
To
that
end,
Rob,
if
you
want
to
collaborate
with
me
on
this,
I
would
like
to
do
that.
I
think
we
can
between
the
two
of
us.
We
can
probably
come
up
with
some
good
docs.
The
second
thing
about
that
is
Ariel
I
believe
internally
at
GitHub.
There's
an
example
of
doing
this
and
detecting
it
rails
console
is
launched
and
not
installing
the
SDK.
Do
you
know
if
there's
a
way
that
that
could
maybe
become
public
knowledge,
because
I,
don't
think,
there's
anything
weird
or
proprietary
about
it?
A
I
just
don't
remember
how
we
did
it,
but
I
remember
it
happening.
That
was
the
first
thing,
but
the
other
broader
thing
was
that
I
think
Shopify
also
has
a
a
sampler
internally.
A
That
like
allows
you
to
apply
a
bunch
of
rules
to
determine
if
a
span
or
a
trace
should
be
sampled.
Something
like
that
is
something
I'm
going
to
see.
A
If
perhaps
we
can
open
source
that,
because
there's
also
nothing
really
proprietary
about
that,
it's
just
a
clever
sampler
that
takes
rules,
and
then
you
can
easily
express
those
rules
and
then
it
does
a
lot
of
magic
and
something
like
that
could
be
very,
very
helpful
to
people
I
think
with
some
documentation
on
how
to
use
it,
like
you
have
a
weird
sampling
need
that,
like
doesn't
fit
into,
you
know
a
probability
based
or
a
head
or
a
tail
based
sample,
or
something
like
that.
A
You
have
a
weird
need:
here's
a
useful
tool
in
how
to
use
it.
So
those
are
my
two
sort
of
thoughts
on
that.
A
B
A
B
I
definitely
I
think
it
actually
dovetails
not
too
terribly
with
at
least
the
semantic
conventions
working
group
discussion,
because
I
feel
like
this
is
something
this
is
like
a
this
has
got
to
be
a
problem
that
is
experienced
in
other
languages
as
well,
where,
like
sometimes
you
want
something
to
Protestant,
or
sometimes
you
want
a
certain
framework
or
a
certain
category
of
instrumentation
to
to
start
traces
and
some
of
them
you
might
not
want
them
to
so
so
how
to
handle
that
I
do
tend
to
agree
that
this
is
probably
something
solved
by
by
a
sampler
and
I
think,
but
I
I
do
think
that
there
hasn't
been
enough
experimentation
with
with
this
or
just
enough
documentation
with
it
I
think
to
be
able
to
easily
just
like
Point
somebody
at
an
exam
at
an
example,
so
actually
implementing
some
of
this
as
a
sampler
or
implementing
or
adding
some
additional
kind
of
fancy.
B
Sampler
like
like
Andrew.
What
was
mentioning
with
some
understandable
rules
based
system
I
think
could
could
go
a
long
way
to
helping
people
quiet
down
instrumentation
of
any
kind
just
by
defining
rules
that
will
kind
of
match
the
spans
that
that
would
be
generated.
B
But
but
I
do
expect
that
with
semantic
conventions,
we're
going
to
get
a
little
bit
more
than
just
like.
This
is
what
the
attribute
should
be
named,
and
this
is
the
format
I
I
do
expect
that
there
will
be
kind
of
some
some
generic
problems
like
in
an
HTTP
instrumentation.
We
want
some
way
to
like
collect
headers
opt-in,
to
like
header
collection,
but
not
all
of
them,
and
not
all
the
time
and
like
how.
B
How
does
that
work
and
what
level
of
uniformity
would
there
be
between
different
languages
and
Frameworks
Etc.
C
C
F
Sounds
like
you
know,
Andrew
has
an
experiment
lined
up
to
say
like
hey.
Maybe
we
can
try
to
use
what
we
have
at
Shopify
and
perhaps
I
need
to
come
around
and
provide
some
more
better
examples
of
how
to
disable
tracing
in
the
rails.
Console
Etc
for
us
and
it
sounds
like
Rob
Ewan
Andrew
are
going
to
pair
up
on
expanding
our
documentation
for
specific
instrumentations
and.
F
A
Because
I'm,
not
a
like
the
prior
art
in
redis,
may
be
useful
and
it's
okay
in
limited
cases,
but
like
I,
don't
think
it's
a
pattern.
I'd
want
to
see
everywhere,
because
I
think
there's
definitely
ways
to
do
it
without
having
to
have
this
option
everywhere,
and
at
least
people
should
be
able
to
like
try
those
and
then,
if
they
don't
work,
we
can
look
at
the
option
like
I.
Think
people
don't
even
know
how
to
do
it
without
an
option
right
now.
A
So,
like
you
should
at
least
make
sure
people
know
first
and
like
then
let
them
try
it
and
see
what
they
think
and
then
involved
based
on
further
feedback,
but
yeah
so
I
would
say
we
I'd,
like
I'm
gonna,
go,
make
a
comment
on
something
about
like
I
think
we
should
hold
off
and
here's
why
and
then
we'll
come
back
to
this
after
we
get
some
better.
You
know.
Thank
you
exactly
something
like
that.
I.
B
And
just
a
closing
thought
that
I
generally
have
in
the
space
is
that
I
don't
know
I
think,
like
the
the
long-term
goal.
I
think
that
that
I
would
have
with
these
things
is
that
instrumentation
should
just
record
the
facts
of
what's
going
on
in
in
a
library
and
if
you
want
to,
if
you
want
to
ignore
some
of
those
facts
or
change
those
facts,
it
would
be
ideal
if
you
could
do
that
somewhere
else.
B
Further
Downstream
in
the
system,
I,
guess
and
not
have
to
like
riddle
the
instrumentation
itself
with
a
bunch
of
books
and
flags
to
do
that
and
I
think
that
might
be
a
little
bit
idealistic
but
like
what
I
have
in
mind
is
doing
some
experiments
with
Samplers
and
really
kind
of
understanding
what
we
can
achieve
there
and
then
also
kind
of
processors.
I
feel
like
I
feel
like
we
aren't
there
today.
I
feel
like
the
processor
pipeline
that
we
have
in
our
SDK
is
is
maybe
could
use
some
redesign
but
I.
B
Think
if
we
be
thought
about
this,
you
can
come
up
with
quite
a
few
ways
to
to
ignore
spans,
remove
data
off
spans.
Annotates
fans
with
some
things
that
that
you
might
want
without
having
to
add
hooks
at
all
by
being
pretty
hands-off
but
I.
B
C
B
B
Yeah,
so
those
are
those
are
my
instincts
anyways
to
want
to
be
able
to
do
things
in
a
hands-off
way
and
I
think
to
tie
this
background
even
to
the
to
other
discussions.
I
feel
like
that
really
will
help
with
like
first
party
adoption
as
well,
because,
like
the
more
Yeah
the.
B
To
add
a
tangled
web
to
your
code
to
support
something
the
less
likely
you
are
to
do
it
and
I
think
you
know
this.
D
C
I
found
myself
saying
something
along
the
lines
of
just
send
it
and
let
the
collector
sort
it
out.
B
B
All
right
so
yeah
I
think
it
seems,
was
a
good
discussion.
Have
we
sufficiently
covered
things?
No
to
the.
C
B
C
To
the
pr
for
now
or
wait
sending
pending
experiments.
B
To
to
experiment
and
yeah
figure
out
ways
to
to
more
elegantly
kind
of
handle
these
situations
going
forward
or
what's
it
what's
possible,
I
guess
in
in
those
realms.
A
Yeah
I'm
gonna
try
to
work
up
a
very,
very
quick
and
dirty
example
of
how
he
could
disable
this
in
the
rails,
console
at
least
so
that
there's
something
to
go
on
and
then
yeah,
Rob
and
I
can
work
up
some
stuff
for
reals,
honestly
like
for
as
much
as
we
talk
about
rails.
Instrumentation
like
our
docs,
are
not
fantastic
about
it
and
I.
A
A
And
also
easy
to
implement
in
one
night,
which
is
the
key,
the
key
advantage
of
why
it
is
that
way
there
you
go,
but
we
I
I
would
personally
be
open
to
redoing
that
in
rails,
if
we
thought
it
was
worthwhile,
it
was
so
easy
and
so
simple
to
do
and
we
needed
a
ruby
option
and
there.
C
D
F
A
F
A
A
A
D
A
Okay,
well
one
one
quick
question:
does
anybody
know
of
any
good
release
Frameworks
in
use
by
other
popular
like
I
know,
rails
has
a
bunch
of
gems.
Do
they
have
a
framework
for
this
that
they
use?
That's,
not
something
horrifically
bespoke
or
are
there
any
other
tools
that
help
do
this
other
than
toys?
A
F
I
mean
that's:
isn't
that
how
the
DD
Trace
RB
is
everything's
in
one
big
old
gem.
F
E
They're
bundled
as
far
as
I
know
still
there
were
some
efforts
too.
That
wasn't
necessarily
like
a
wonderful
pattern
that
people
loved
it
never
got.
You
know
negative
feedback
on
so
I,
actually
don't
know
what
the
state
or
the
union
is.
There.
E
Time
it
was
it
made
stuff
easier
by
March.
F
D
C
So
there's
a
thing:
we're
currently
cobbling
together
a
python
Hotel
Bistro
and
a
thing
that
I
like
about
their
pythons
python
hotel
has
a
package
called
open,
telemetry
instrumentation.
C
It
is
a
package
that
depends.
It
is
a
standalone
package
and
it
has
within
it
two
commands
it
might
have
more
than
this.
But
the
two
things
that
I
find
interesting
is
a
command
called
open,
Telemetry
bootstrap,
which
evaluates
your
current,
your
app
this
particular
application's
current
set
of
dependencies
and
goes
and
injects
what
it
knows.
C
Instrumentations
exist
for
so
instead
of
it's
like
the
equivalent
of
all.
But
it's
not
I
have
a
direct
dependency
on
everything.
It's
I
just
am
aware
of
everything's
existence
and
I
will
look
at
your
dependency,
your
the
dependencies
that
you
have
and
I
will
add
to
them.
The
instrumentations
for
the
things
that
you
use
so
the
result
of
running
an
open,
Telemetry
bootstrap
is
a
requirements
text
or
whatever
updated
with
the
instrumentations
for
the
things
that
you
use
and
not
everything,
and
all
that
we
have
in
all
that's
pretty
dope.
C
But
yes,
but
a
pushback
I
got
say
as
an
example
from
the
Mastodon
folks
was
like
they
looked
at
the
gem
file
lock
after
bringing
it
open
to
our
open
Telemetry
all
and
they
were
like
that's
a
lot
of
dependencies,
like
half
of
which
are
used
right.
A
C
A
It's
like
a
it's
like
a
setup
task.
You
have
to
run
it
just
tells
you
what
you
should
install.
It
doesn't
actually
install
anything
itself.
Yeah
I've
used
it.
It
was
kind
of
nifty.
I
could
see
why
people
might
like
that
right.
C
D
C
Next
month,
next
year,
it's
like
maybe
we
write
a
so.
C
Don't
have
to
use
at
all,
but
yeah
you
as
a
human,
have
to
go
and
Survey
what
information
instrumentations
are
available
and
do
I
use
it
and
well
like
one
of
the
beauties
of
all
or
the
well
either.
Either
pattern
using
our
all
or
their
bootstrap
detects,
what's
instrumented
and
bring
them
in
is
that
if
I'm
totally
inexperienced
with
an
app,
let's
call
it
Mastodon
I've,
never
run
it.
I
know
it's
rails,
app
I'd
drop
one
of
these
things
in
there
and
it
gets
lit
up
and
I.
C
A
D
F
All
different
ideas
right,
so
the
one
one
idea
as
to
is
to
generate
the
subset
of
dependencies
in
your
gem
file,
yeah
based
on
the
dependencies
that
you
already
have
another
one,
is
to
I'm
going
kind
of
make
the
all
gem
not
have
transitive
dependencies
and
roll
everything
up.
I'm.
F
It
could
be
the
it
could
be.
The
maven
jar,
yeah
I
have
the
other
option
of
like
steer
people
towards
you
know
cherry
picking
what
instrumentations
they
want
and
not
use
all
by
default,
all
those
have
their
drawbacks.
Yeah.
D
F
Script
had
to
have
some
sort
of
hooks
into
bundler,
so
it
has
some
sort
of
detection,
and
the
other
thing
that
we
haven't
talked
about
is
kind
of
like
sneaking
instrumentations
in
without
explicitly
declaring
them
so
like
rapping
bundle,
exact
and
adding
a
dash
R.
You
know
like
to
Ruby
adding
a
dash
required
Library,
which.
C
C
A
You
know
about
a
bundler
plug-in
a
thing
that
I
just
learned
about
yesterday,
which
enhances
the
function
that
I
know
right.
I
saw
someone
write
like
a
plug-in
in
an
internal
commit
and
I
was
like
that
looks
like
you
copied
and
pasted
a
very
old
guy
that
can't
be
right,
but
then
I
looked
and
no
it's
a
thing.
It's
still
a
thing.
A
It's
a
way
to
do
additional
bundler
commands,
so
we
could
have
like
a
bundler
plug-in
that,
like
is
like
bundle,
Auto
instrument
and
then
that
can
be
part
of
people's
I.
Don't
know
that
can
be
like
how
we
do
the
that.
C
A
Can
bundler
could
tell
you,
maybe
it
even
hooks
into
like
bundle,
outdated
or
something,
and
it
says
FYI,
there's
a
new
instrumentation
for
you
or
something
like
that.
I
mean
I,
don't
I,
don't
know
how
far
I'm
making
stuff
up
but
yeah.
That's
a
potential.
That'd
be
fun,
but
anyways
I
have
a
hard
stuff.
So
I
will
see
you
all
later.