►
From YouTube: 2022-09-20 meeting
Description
cncf-opentelemetry meeting-2's Personal Meeting Room
A
A
A
All
right
so
I
think
we're
gonna
be
Shopify
free
today,
since
they've
got
like
some
sort
of
thing
going
on
and
so
I
guess
of
our
regulars
I
think
the
only
other
person
would
be
Rob
kid,
so
I
mean
effectively
I
think
between
Nicholas's
iPhone
myself
and
yourself.
We
have
a
quorum.
B
Yeah
yeah,
it
sounds
like
we
have
a
quorum,
so
let's
go
ahead
and
get
started.
If
anybody
else
comes
in
they
can
we
can
catch
them
off.
B
This
first
issue
this
first
issue-
I,
don't
know
I,
don't
know
about
this
thing.
This
thing
was
bringing
back
like
a
lot
of
horrible
memories.
When
people
were
talking
about
it
and
basically
yeah
this.
The
author
of
this
VR
wants
to
be
able
to
fail
fast
for
evalid
genome
values.
B
B
So
the
author
was
not
really
here
for
the
conversation
to
fully
continue,
but
yeah
I
don't
know
Ariel,
it
seemed
like
you
were
maybe
having
some
some
thoughts
as
I
was
rambling
about
this
nah.
A
It's
just
the
idea
of
failing
fast
on
enum
values
is
like,
then.
Your
deployment
is
completely
broken
in
situations
which
I
I
feel
like
I
feel
like
that's,
like
a
compatibility.
B
A
B
Yeah
and
I
think
this
must
really
be
coming
from,
like
in
Death
Mode
scenario,
where
the
user
was
trying
to
set
stuff
up
and
which
is
really
confused.
That
you
know
was
maybe
passing
in
an
invalid
value
for
for
some
configuration-
and
you
know,
was
not
getting
the
feedback
that
they
expected,
but
yeah
I.
Think
in
in
a
production
scenario
like
you
want
your
services
to
start,
you
don't
want
them
crash
looping,
just
because
a
config
value
is
wrong.
B
So
anyways
I
don't
think
there
was
any
resolution.
It
does
look
like.
There
is
a
couple
of
approvals
on
this,
but
I
think
there
will
be
some
further
discussion.
Foreign.
B
B
B
We
talked
about
this
a
little
bit.
It
was
just
kind
of
being
surfaced
again
at
the
specsig,
but
mainly
just
proposing
that
otlp
exporters
add
some
kind
of
user
agent
header
to
identify
where
the
Telemetry
is
coming
from
and
I.
Don't
think
anybody
disagrees
with
the
idea.
I
think
there's
maybe
a
little
bit
like
bike
shedding
about
exactly
the
structure
of
of
the
string,
but
it
looks
like
it's
getting
approvals,
so
it's
likely
to
merge
soon.
B
B
There
is
a
PR
to
change
the
default
buckets
for
the
explicit
histogram,
and
there
was
some
debate
about
whether
or
not
this
was
would
be
a
breaking
change.
Basically,
your
old
buckets
were
these
blah
blah
and
they
go
up
to
1
000
and
new
buckets
would
be
these
going
up
to
ten
thousand
so
yeah
there
were
some
debates
as
well
as
or
not.
B
This
would
be
breaking
and
I
think
there
were
some
good
arguments
that
it's
not
actually
breaking
to
add
new
buckets
in
most
of
your
or
I
think,
like
I
said,
would
depend
on
your
back
end,
but
most
most
back-ends
should
the
querying
should
not
change
due
to
like,
essentially
just
maybe
some
slight
slightly
better
resolution
on
the
buckets
or
additional
buckets
as
you.
B
B
B
There
has
been
some
work
in
progress
to
add
some
new
process.
Metrics
I
think
there
were
some
spec
issues,
maybe
directions
I'm,
not
sure
there
was
a
direction
attribute,
kerfuffle,
I,
don't
know
if
that
was
what
was
holding
these
back.
I
know
something
spec
wise
was
kind
of
just
holding
this
back
from
from
being
accepted,
but
the
most
whatever
the
issue
was,
was
resolved
in
the
most
recent
spec
release,
and
it
looks
like
there's
quite
a
few
approvals.
B
There
are
some
semantic
conventions
for
some
browser.
B
I,
don't
think
there
was
any
any
objections,
they're
all
fairly
new,
so
more
metrics
are
coming.
B
B
B
B
B
Appears
to
be
somewhat
weird,
maybe
somebody
could
explain
it
to
me,
but
you're
supposed
to
send
a
open,
Telemetry
scope
info
metric
that
if
you're
a
Prometheus
receiver,
you
would
you
would
receive
that
metric.
But
you
would
not
actually
export
that
to
Prometheus.
You
would
use
that
to
figure
out
the
scope,
name
and
version
that
you
want
to
stick
in
otlp
for
that
batch,
but.
B
B
B
Yeah
I
think
that's
what
this
one
this
particular
part
is
about
and.
B
C
B
C
C
Yeah
or
I
was
trying
to
figure
out
if
this
is
like
within,
like
the
use
of
an
SDK,
if
you've
got
Prometheus
around
him
like
Prometheus
is
generating
but
you're
going
to
export
it.
But
this
is,
you
know
and
I'll,
go
read
this
PR
and
see
if
it's
just
scope
to
overload
a
word.
If
this
only
pertains
to
collector
Transformations
or
it
goes
beyond
that.
B
B
C
B
B
Yeah,
it
looks
like
it,
it
needs
approvals
to
signify
the
happiness,
but
that's
that's
where
this
thing
has
landed.
B
And
then,
lastly,
but
not
leastly,
there's
just
a
question
of
whether
or
not
the
exponential
histograms
should
be
moved
to
stable,
I.
Think
I
think
this
is
more
of
a
question.
It
should
be
stable
aspect
level
and
apparently
there
are
implementations,
in.net
go
Java
and
then
python
has
an
open
PR.
B
So
there
was
a
little
bit
of
debate
going
on
about
or
the
debate
that
was
going
on
was
coming,
maybe
from
the
Java
world
and
I
think
they
were
they're,
not
very
happy
with
their
implementation
of
the
exponential
histogram
and
it
seems
to
be
for
the
costly
in
a
performance
performance,
wise,
so
I
think
one
suggestion
was.
We
would
like
to
better
understand
when
to
recommend
an
exponential,
Instagram
or
a
fixed
bucket
and
when
you're
actually
winning
but
having
one
over
the
other.
B
But
then
the
the
discussion
kind
of
was.
B
I
don't
know
the
discussion
changed
in
that.
Maybe
it's
just
the
Java
implementation
is,
is
actually
the
core
problem
there
and
that
an
exponential
histogram
should
really
not
have
any
significant
overhead
over
a
fixed
bucket,
histogram
and
I.
Guess
the
underlying
data
structure
that
Java
is
using.
They
were
at
least
saying
that
yeah,
it
is
probably
not
the
right
data
structure
and
probably
it
caused
a
lot
of
the
issues
there.
B
So
after
that
discussion,
I
think
people
were
more
ready
to
say
you
know
this
is
probably
stable
from
a
spec
perspective
and
languages
can
refine
their
their
implementations
as
needed
to
kind
of
resolve
any
performance
issues.
C
B
But
excellent:
do
we
have
any
issues,
PRS,
burning
questions.
B
Do
we
need
to
release
instrumentation
at
all?
I
saw
some
instrumentation
did
get
released
already,
so
maybe
we're
yeah.
A
A
A
So
there
was
also
a
discussion
that
someone
opened
up
specifically
around
what
the
names
of
the
spans
are
right.
Now.
What
it
uses
is
the
controller
name
and
the
action
and
I
guess
the
action
to
name
the
span
so
it'll
be
anything
like
you
know:
users,
controller,
octothorpe
and
then
whatever
index,
and
this
is
looking
to
see
if
we
can
use
the
rails
route
instead,.
A
Yeah
I
think
it's
it's
a
combination
of
both
right,
it's
a
more
useful
name,
I,
think
and
and
some
comp
and
then
I
asked
the
question
and
I
want
to
hear
back
from
Robert
when
he
comes
back.
I
think
he
was
the
original
author
of
this
is
essentially
for
us
to
match
these
up.
I
think
a
little
more
clearly
with
what
I
think
would
match
up
with
zemcon.
So
I
agree
that,
for
example,
the
rails
route
should
probably
then
be
the
name
of
the
span.
A
A
A
So
it's
a
so
now
we're
kind
of
like
wondering.
What's
gonna,
be
the
best
outcome
here,
because
it
could
be
the
case
that,
like
we
have
a
naming
strategy
for
the
span
so
we're
either
keeping
what
Iraq
gave
it
overriding
it
with
what
rails
resolved
it
to
in
a
templated
route,
naming
it
controller
pound,
name
or
whatever
right,
I
use,
octo,
Thorp
and
pound,
so
I'm
going
to
throw
a
hashtag
in
there
for
the
next
one
and
I'm
curious
that
I've
got
two
vendors
on
the
phone.
C
I
think
I
need
to
read
it
more
deeply
in
general,
without
having
read
this
deeply
I
would
say
like
the
name
of
the
span:
autofollow
sencom.
So
if
it's
HTTP
verb
and
Route
as
the
name,
we
should
consider
making
that
the
name
of
the
span
and
then
all
the
other
attributes
that
sort
of
apply
are
good
to
add.
So,
in
addition
to
http
verb
and
Route
in
the
name,
there's
HTTP
route
field
that
would
have
the
route
on
it
and
well.
C
B
Yeah,
if
I
mean
I
would
say
from
my
perspective,
I,
don't
think
it
really
matters,
but
the
having
things
be
be
uniform
with
semantic
conventions.
I
think
you
know
having
the
HTTP
method
and
Route
is,
is
as
good
as
any
name
for
a
span
and
as
long
as
the
rest
of
the
data
is
there
as
attributes
and
I,
think
everybody
can
I
think
everybody
can
query
the
data
you
know
in
in
a
sane
and
reasonable
way,
but
yeah
I
would
I
would
turn
that
question
back
onto
you
as
a
user.
B
A
Doesn't
because
we
don't
use
this
instrumentation
specifically
because
our
previous
open
trading
implementation
manually
name
the
HTTP
method,
slash
whatever
it
resolved
as
the
route
path.
Essentially,
and
so
you
know,
the
concern
that
comes
up
is:
is
the
rails
routing
system,
since
the
route
itself
is
not
available
to
you
at
the
time
that
this
block
of
code
is
executing
you're
Computing
the
route
multiple
times
in
a
single
request,
and
it
would
be
once
right,
but
is
there
some
concern
about
overhead
for
computing
a
specific
route
multiple
times.
A
A
The
resolved
route
is
not
available
and
you'll
also
note
that
the
resolve
route
doesn't
necessarily
match
what
some
cons
would
say
right
so,
like
you
could
see
that
it
says
like
yeah,
okay
and
then
it
has
the
suffix
of
the
format.
If
you
look
at
the
test
case
line
28
on
metal
test-
and
you
imagine
that
if
this
is
a
resource
route,
it's
going
to
have
the
parameterized
route
with
colons
in
it
and
I
think
that
some
conf
is
more
like
hey
use,
curly
braces
or
whatever
to
the
Limit
parameters
in
this.
A
But
unable
recognized
route
is
disabled
by
the
Fall
and
what
we
don't
have
is
like
a
paper
trail
of
why
I
was
disabled
by
default,
which
is
you
know,
adding
that
route
attribute
as
well.
So
this
is
where
my
question
comes
from.
Oh,
is
this
a
problem
that
we
compute
the
route
every
time
Prairie
request
I.
B
Yeah
I
think
I'm
not
sure
how
how
fancy
rails
routing
is
here.
If
there
is
like
some
cash.
No,
it's
it's
gonna.
It's
gonna
go
through
all
of
the
all
the
work.
A
Underneath
that,
underneath
the
the
data
structure
that
it
has
in,
there
is
essentially
like
an
AST,
so
you
know
you're
crawling
through.
They
asked
you
trying
to
find
something
that
matches
and
I.
Don't
know
why
there
would
be
multiple
matches,
but
recognize
yields
a
block
for
every
match.
It
finds
foreign.
C
Since
this
instrumentation
is
dependent
on
our
rack
instrumentation,
could
we
in
our
Iraq
instrumentation
make
it
a
feature
of
rack
instrument
tape?
No,
this
is
a
rails,
not
arachna.
Oh
okay,.
B
We
could
pull
requests
into
rails
that
you
stash
the
resolve
route
in
the
end
and
then
a
decade
from
now
we
can
rely
on
it
being
there.
A
There's
that
part,
but
I
would,
in
the
case
where
we're
saying
like
we
want
to
add
things
to
the
to,
like
the
rack
environment,
I
would
hate
to
add
a
thing
like.
Oh
here's,
the
official
rails
route,
but
it's
only
available
to
you.
If
you
know
you
used
our
instrumentation
and
then
somebody
shut
off
that
instrumentation
or
a
day
and
use
it,
and
they
were
relying
on
that
value
being
present
in
the
environment.
That'd
be
weird.
A
So
if
it
feels
like
it
would
be
more
of
like
a
context
attribute,
as
opposed
to
rack
request
attribute
if
we
were
to
add
it
from
instrumentation
and
make
it
available.
Downstream.
C
Yeah
I
didn't
yeah,
it
wasn't
a
well
sought
out
idea,
apparent
just
hey.
A
Rob,
you
know
I'm
I'm
here
about
I'm
all
for
it
well
thought
out
or
not.
C
B
C
C
B
B
C
Will
I
will
read
this
one
to
Ariel
I'm
a
fan
of
getting
this?
The
thing
that
comes
out
of
this
instrumentation
to
conform,
the
semantic
conventions
and
there's
performance
questions
about
Computing
the
route
multiple
times
so
like.
If
we
could
figure
out
an
implementation
that
reduces
the
number
of
times
the
route
is
resolved
so
much
better
right
that
the
I.
A
Think
that
summarizes
that
summarizes
it.
But
again,
I
want
to
hear
from
the
reasoning
behind
like
enabling
the
route
as
being.
A
Route
attribute
that
really
was
for
performance
concerns,
because
I
couldn't
find
a
paper
trail
for
it.
A
So
we
can
all
we,
you
know
we're
already
adding
the
route
and
Computing
the
route
as
long
as
enable
recognize
route
is
configured
and
that's
off
by
default.
So
the
question
is:
was
it
off
by
default?
This
is
my
confirmation.
Why
was
it
off
by
default?
Was
it
off
by
default
because
of
performance
issues?
I
see,
you
know
that
might
give
us
a
little
bit
of
indicator
or.
A
Having
a
different
shape,
it's
it's
unclear
to
me.
It's
like
at
least
I
didn't
find
anything
that.
Let
me
just
get
me,
maybe
my
my
my
archeology,
my
kid
archeology
isn't
working
well.
B
I
will
say
that
I
have
seen
you
know,
controller
controller
in
action
as
the
name
for
spans
as
a
popular
naming
strategy,
because
I
think
it's
easy
or
is
it
it's
an
easy
thing
that
makes
sense
in
the
rails
world
some
vendors
I
know
go
with
that
name
and
one
particular
vendor
that
uses
that
name.
It's
like
the
Ruby
ecosystem
is
kind
of
a.
B
Everything
else
actually
uses
the
verb,
HTTP
verb
and
Route,
so
I
think
the
last
the
less
we
can
be
a
special
snowflake
and
probably
still
get
not
at
a
huge
performance
penalty
at
least
would
be
ideal.
C
C
B
Cool
I
feel
like
this
was
a
good
conversation.
I
feel
like
we
can't
really
complete
it
without
80
of
Shopify
in
attendance,
yeah
yeah.
We.
A
A
Yeah
your
feedback
and
input,
though
on
the
in
this
conversation,
would
be
very
helpful
if
you
could
write
them
up.
Even
though
we've
we've
discussed
this
a
bit
in
hold
on
and
looks
like,
we
have
a
cat
coming
to
give
us
a
demo.
We
do
he's.
B
B
C
A
Oh
boy,
let's
see.
A
A
Mean
I
see
the
race
car
was
approved.
Maybe
we
just
need
to
merge
that
one
and
get
a
release
out
for
it
that
Force
flush
option
looks
like
it
was
approved,
so
we're
at
that
point
of
merging
those
two.
So
I
have
no
idea
why
we
didn't
just
merge
it
was
there
Nick?
Was
there
anything
yeah.
C
C
B
C
Yeah
I'm
very
unfamiliar
with
green
three
minutes.
It
just
feels
like
there's
a
very
specific
Ruby
3.1
issue.
C
This
is
something
not
very
new,
so
it
doesn't
detect
some
time
and
figure
out
how
we
can
get
that
going,
but
I
actually
think
this
Ariel
mentioned
it's
that
we
actually
don't
use
mini
test
once
the
standards
to
kind
of
the
horseback
marks
instead,
so
I
was
thinking
the
alternative
words
just
to
add
that
and
replace
the
related
stuff,
because
there's
only
actually
two
specs
I
think
it's
in
the
red
side.
B
Okay
yeah,
it
sounds
like
you
at
least,
have
some
some
ideas
here.
B
C
I'm
intrigued
by
the
the
pull
request
for
untraced
endpoints
to
take
a
regex.
C
I
know
we
have
users
who,
like
filtering
out
the
instrumentation
and
I,
can
see
how
our
rejects
would
be
useful.
I
worry
about
taking,
as
Andrew
has
mentioned,
worrying
about
taking
a
regex
from
the
environment.
C
C
But
I'll
read
this
PR
and
pine:
there
yeah.