►
From YouTube: 2023-02-28 meeting
Description
cncf-opentelemetry meeting-2's Personal Meeting Room
A
A
A
A
A
So
here
is
here's
one
of
the
novels.
Maybe
this
one
is
not
quite
so
bad.
There's
a
lot
of
comments
in
there.
It's
at
least
trying
to
provide
clear
guidance
on
what
semantic
conventions
actually
enforces.
A
So
what
is
it
in
force
and
what
is
it
not
enforce?
Just
so
that
that
is
clear,
definitely
enforce
the
attribute
key
names.
Pipes
span
names,
metric
names
units,
but
there's
this
Clause
to
possibly
expand,
allowed
changes
to
semantic
conventions
to
allow
metric
attributes
that
would
not
increase
Time
series.
Cardinality
Etc
I
think
there
is
there's
also
some
discussion
about
enums.
A
If
you
can
add
to
like
an
even
list,
I
think
was
another
thing
I
saw
in
here.
At
any
rate,
this
is
trying
to
Define.
What's
What's
inbounds
for
for
semantic
conventions
feel
free
to
read,
comment.
A
We
talked
a
little
bit
about
this
for
metric
semantic
conventions
wanting
to
align
on
seconds
as
the
units
mainly
to
align
with
Prometheus
I.
Think
people
are
getting
serious
about
this
and
asking
for,
like
last
call
so
Riley
was
saying
we
will
make
the
change
which
aligns
with
Prometheus
as
decided
at
the
Tito's
Valentine's
Day
edition
of
the
specsig.
A
Oh
and
then
I
don't
think
we
totally
got
into
the
details
of
some
of
the
other
HTTP
semantic
conventions
stuff
that
is
on
the
agenda,
discovers
so
much
time
on
these
other
things,
but
there
is
still
kind
of
the
HTTP
signing
conventions
group.
They
are.
A
I,
don't
know,
there's
this
issue
about
confirming
that
it's
okay
to
Mark,
http's
semantic
conventions
stable
without
marking
the
net
attribute,
stable,
I,
think
I
think
they
are
just
trying
to
get
feedback
on
this
I.
Don't
know
that
they
actually
think
that
the
HTTP
attributes
are
totally
stable
yet
because
they
still
have
some.
A
That
they're
asking
people
to
look
at
but
I
think
maybe
they're
projecting
into
the
future
trying
to
kind
of
divide
up
their
work
that
the
net
dot
attributes
are
not
not
going
to
be
blockers
when
they
are
ready.
A
They're
all
just
different
layers
of
the
same
thing
or
something,
but
this
will
be
yet
another
breaking
change,
ACP,
client,
IP
to
http.4.4.
A
And
then
introduce
a
forwarded.proto
attribute
one
thing
that
did
kind
of
come
up.
That
is
not
on
the
agenda
at
all.
Is
that
I
don't
know?
There's
HTTP
scientific
inventions
working
group
and
a
lot
of
the
changes
they
are
making
are
kind
of
aligning
a
little
bit
more
through
as
a
common
schema.
A
And
then
there
was
somebody
from
elastic
common
schema
and
I
noticed
that
there
is
yet
an
that.
A
an
Otep
in
the
past
was
an
otap
I
think
to
adopt
elastic
common
schema
and
then
the
elastic
folks
got
busy
and
it
kind
of
disappeared
or
lost
Focus.
So
they
opened
another
one
in
December
and
it
seems
like
they
are
having
some
internal
discussions
they're.
A
Having
some
discussions
with
some
folks
in
hotel
about
trying
to
adopt
ECS
and
I
think
there
would
be
some
kind
of
some
kind
of
merge
process,
I
guess
where
they
would
kind
of
look
at
what
already
exists
in
a
hotel
and
kind
of
at
least
negotiate
as
to
like,
when
both
have
specified
the
same
thing
like
which
one
they
should
actually
use.
A
But
but
yeah
I
know
T
grin
kind
of
came
out
and
said
he
did
not
does
not
support
this
o-type
in
the
current
form,
but
he
supports
the
idea
of
converging
to
ECS.
So.
B
B
A
Yeah
so
I'm
not
sure
I'm,
not
sure
what's
happening,
but
ultimately
it
does
seem
like
if
we
did
adopt
ECS.
That
could
possibly
be
a
shortcut
through
semantic
conventions.
If,
if
the
conventions
are
okay,
that's
my
two
cents.
A
Oh
yeah
I
did
end
up
in
the
agenda
I
believe
this
person
was
from
Amazon
and
kind
of
interested
in
the
context.
Scoped
attributes.
A
Think
you
can
carry
on
with
it
definitely
yeah.
She
was
very
interested
she's
like
how.
How
soon
can
we
have
this,
and
then
everybody
told
her-
probably
never
not
actually,
but
the
like
we're
really
busy
right
now
and
people
have
been
wanting
this
for
a
while.
It's
funny
it
will
probably
take
a
lot
of
work
to
get
across
the
finish
line,
but
I
don't
know
to
me.
A
This
seems
like
it
does
need
to
be
a
focus
at
some
point
sooner
than
later,
because
people
keep
asking
for
it
and
I,
don't
think
they're
gonna
stop
asking
for
him,
so
we
need
to
find
a
way
to
make
this
happen.
Someday.
C
A
A
This
is
the
configuration
Otep
and
it's
200
comments.
100
comments,
sorry
exaggerating
here,
but
ultimately
there
is
a
group.
That's
been
working
on
kind
of
a
configuration
file
format,
at
least.
A
From
what
they
were
saying
like,
basically,
you
would
use
a
file
or
environment
variables,
but
there's
no.
These
things
do
not
work
together,
at
least
not
in
their
current
forms.
There's
no
kind
of
like
you
can
provide
the
file
and
then
override
with
an
environment
variable,
or
vice
versa,
which
I
thought
would
be
possibly
problematic.
The.
A
I
think
the
one
thing
that
you
can
do
is
you
can
provide
the
path
to
the
configuration
file
via
environment
variable,
but
then
it
would
just
use
the
file
and
then
I
do
think
they
need
to
like
loosen
that
a
little
bit
for
some
like
secrets
that
will
need
to
be
injected
by
an
environment
variable
so
I
think
there
will
be
some
special
casing.
A
Perhaps,
but
I
definitely
did
not
read
through
this
whole
thing,
and
it's
and
the
comments,
but
I
was
listening
to
the
summary
and
those
were
things
I
was
taking
away
from
the
summary.
I
did
kind
of
spy
these
code,
samples
of
kind
of
how
it
would
look
no.
A
I
think
there
is
yeah,
so
the
other
thing
is
that
this
is
based
on
Json
schema.
So
there
is
actually
a
schema
for
the
configuration
file.
I
suspect
that
this
will
be
a
huge
nightmare
actually
and
probably
needs
to
be
looked
over
to
figure
out
like
how
we
provide
instrumentation
config,
because
I'm
pretty
sure
they
didn't
schema
out
how
that's
going
to
look
for
Ruby
or
for
for
any
other
language.
So
these
might
be
things
that
we
need
to
kind
of
figure
out.
A
How
that
will
high-end
so
use
this
Json
schema
but
and
I
think
it
requires
that
you
support
at
least
Json,
but
can
support
yaml
and
there's
some
way
that
you
can
convert
your
yaml
to
Json
and
run
it
through
a
Json
schema
and
or
there
is
some
pipeline.
Maybe
it's
the
other
way
around
that
you,
you
can
transform
between
Json
and
yaml
and
still
validate
your
yaml
via
Json
schema.
B
A
It
survived,
Andrew
is
mentioning
that
Json
is
a
subset
of
yemo,
so
that's
That's
the
basis
for
how
this
magic
works.
A
A
A
What's
that
any
questions
comments
concerns
with
the.
A
Something
consider
having
some
friends
and
sharing
the
titles.
A
D
These
questions
directed
at
yourself
and
Francis
who's
not
present,
but
I'm,
trying
to
understand
kind
of
like
the
HTTP
proxy
span
itself.
That's
always
like
you
know
kind
of
like
re-implementing,
this
rack
middleware
using
the
event
API.
D
There
was
some
stuff
that
you
know.
I
was
trying
to
understand
some
features
that
existed
in
the
existing
middleware
to
try
to
bring
that
parody,
and
so
a
couple
things
that
came
to
mind
when
I
was
looking
at
this
right
is
the
A2
proxy.
Expand
itself
is
the
as
a
server
type,
and
so
what
happens
is
if
somebody
enables
this.
It
adds
that
one
HTTP
proxy
span
that
represents
the
difference.
D
You
know
the
request
start
time
to
some
endpoint,
which
I
believe
is
the
end
processing
time
of
that
server
request,
whether
you
it'll
have
the
same
end
time
stamp,
but
it
doesn't
kind
of
calculate
the
total
time
kind
of
like,
including
the
response,
bytes
and
everything
you
know
for
that
proxy,
because
that
intermediary
is
gonna,
do
some
post-processing
or
the
the
stream
response
right.
So
it's
kind
of
a
fuzzy
representation
of
like
trying
to
give
you
what
looked
like,
potentially
the
queue
time.
D
You
know
like
the
receive
time
of
the
proxy
and
the
queue
time
in
that
proxy,
but
get
to
know
other
attributes
all
of
the
attributes
that
you
would
normally
see
on
a
server
span.
As
far
as
I
can
tell
like
the
HTTP
status
all
end
up
on
the
internal
raxman.
D
D
B
To
get
my
head
around
the
situation,
I'm
gonna,
try-
and
these
are
one
of
my
Infamous
questions
in
the
form
of
a
statement
this
proxy.
This
span
is
created
to
represent
the
the
activity
in
some
like
front-end
load,
balancer,
ahead
of
whatever
rack
app
is
running
yeah.
It's.
B
It's
it's
it's
a
different
process
and
not
the
Ruby
process
that
has
received
the
request
right.
It's
like
we're.
Creating
this
ghost
span
that
that
Ruby
kind
of
can
know
existed
because
there's
a
proxy
ahead
of
it.
B
In
my
mind,
I
would
expect,
if
rack
is
making
this
like
implied
span
ahead
of
it
in
the
trace,
that
that
would
be
a
server
and
then
it
would
continue
to
create
a
service
band
for
itself
representing
because
this
was
even
though
it's
these
spans
are
being
generated
by
this
single
process.
The
this
proxy
spans
representing
a
a
different
process
ahead
of
the
rack
service,
so
I
would
expect
them
both
to
have
server
on
them,
because
this
proxy
span
represents
the
initial
intake
of
something
else.
D
I
agree,
I
understand
what
you're
saying,
however,
the
other
links
that
I
supplied
there
are
for
unresolved
discussions
around
how
we
should
represent
proxy
spans.
Okay,
because
you
know
the
specification
right
now
is
kind
of
like
saying
hey,
you
know
we
need
to
have
once
banned
out
as
a
client
one
Spanish-
and
you
know
this
has
happened
multiple
times,
it's
kind
of
like.
D
What's
actually
the
client
span
and
what's
actually
the
service
fan
and
does
it
matter
that
we
have
more
than
one
and
there's
always
this
desire
to,
like
you
know,
have
the
one-to-one.
There
has
been
some
discussion
but
I
feel
like,
but
there
is
no
resolution
about
how
we
represent
these
intermediary
spans
right,
because
it's
not
just
it's
any
instrumentation
that
occurs.
That
is
not
the
actual.
Like
I,
don't
know
code
under
your
service,
I
guess
under
your
purview,
I,
don't
I,
don't
know
how
to
what
do.
D
That's
related
to
it
is
going
to
be
a
server
spam
and
there's
a
one-to-one
correlation
to
those
and
there's
been
discussions
about
having
internal
clients
and
this
and
that,
but
it's
kind
of
like,
depending
on
how
deep
the
stack
goes
if
you're,
like
writing
traces
for
like
freaking
soccer
calls
or
whatever
which
one's
the
client
yeah
backwards,
and
it
doesn't
matter
that
you
have
multiple
clients
and
so
in
as
part
of
the
discussion
was
kind
of,
like
somebody
said,
I
think
it
was
the
person
I.
A
C
B
C
D
But
if
you
are
a
program
who's
trying
to
represent
istio,
because
you
don't
have
the
ability
to
instrument
it
or
you
know,
you're
in
AWS
land
or
whatever
and
you're
like
or
you're,
using
nginx
and
this
time
not
to
install
open
telemetry
module
on
it
right,
like
is
the
applications.
The
thing
should,
the
instrumentation
of
your
application
be
responsible
for
doing
that.
Representation,
yeah.
A
So
yeah
I
I,
don't
know
just
to
chime
in
with
a
couple
of
things.
I
do
think
this
is
important
data
I
do
remember
like
yes,
this
is
ultimately
Q
time.
The
way
it
really
works
is
that
the
thing
that
receives
requests
sets
a
header
with
a
time
stamp
and
then,
when
you
actually
process
the
request,
you
can
do
the
subtraction
and
figure
out
how
long
this
thing
was
like
waiting
to
be
processed
and
I.
A
Think
there
have
been
some
semi
famous
cases
where
people's
like
instrumentation,
like
their
requests,
are
processing
very
quickly.
Everything
looks
totally
fine
from
the
Telemetry,
because
you
don't
realize
that
the
thing
was
queued
for
10
seconds
before
the
quick
request
was
processed,
so
I
think
yeah
I,
my
two
cents
is
that
I
don't
know
it's
like.
A
D
Well,
yeah
I
I
mean
and
that's
and
that's
essentially
the
allowed
attribute
header
X
request
started.
So
if
you
add
that
to
your
allowed
list,
then
it
shows
up
as
an
attribute
and
then
but
then
that
requires
the
vendor
to
kind
of
like
do
some
sort
of
specific
parsing.
As
far
as
the
event
is
concerned,
what's
weird
about
the
event
is
that
the
time
stamp
drift
occurs
where
it's
like
the
event
occurred
before
the
span
was
started
and.
C
D
But
anyway,
I
think
I
think
what
I'm
getting
at
is
I
found
the
HTTP
proxy
span
a
bit
confusing
and
for
it
not
to
have
the
same
attributes
for
it
not
to
a
appear
to
me
to
be
a
server
span
with
what
I
would
expect
server
spans
to
have
on
them
like
the
HTTP
response,
like
the
HTTP,
Target
and
so
on
and
so
forth.
Ad
nauseum
I
found
that
to
be
unusual.
A
Yeah
I
think.
A
A
I
do
think
that
the
this
there
is
the
HTTP
semantic
conventions
working
group
and
I
feel
like
this
is
exactly
in
their
wheelhouse
and
I
was
consider
trying
to
like
at
mention
Ed
mentioned
and
have
some
conversations
with
the
HTTP
semantic
conventions
working
group
and
see
if
they
agree
that
this
is
kind
of
something
that
should
be
on
their
roadmap
and
at
least
I.
D
I
guess
for
me,
it
means
like
do
we
continue
to
keep
this
functionality
if
I
don't
understand
it
and
I
can't
really
explain
it
to
people
and
how
they
should
use
it
and
what
value
they're
going
to
get
out
of
it?
That's
you
know
that's
kind
of
like
my
bit
about
it
like,
like
you
know,
I
also
ran
into
another
thing
where
it's
like
there's
some
functionality
got
ported
but
incomplete,
so
I
open
a
PR
to
like
remove
it
like
little
things
like
that.
D
So
these
are
the
things
that
I'm
kind
of
like
on
the
lookout
for
when
I'm
and
that's
why
I
wanted
to
again
kind
of
like
rehash
this
conversation
with
you
since
I
know
that
you
Francis
and
the
original
author
had
had
discussed
this,
and
this
was
like
an
import
I
guess
this
was
a
component
that
was
available
in
DD
Trace,
and
so
that's
why
it's
like
it
was
eligible
for
being
pulled
in
and
I
I
could
see.
The
value
I
could
see
why
people
want
it.
D
It's
like,
if
you're
deployed
to
like
a
managed
service
like
Heroku.
It's
like,
oh
yeah,
that'd,
be
really
nice
to
see
that
or
if
you
can
instrument
nginx,
it
would
be
nice
to
see
that
represented
somehow,
but
it
just
seems
ambiguous
to
me
about
like
if
it's
adding
value
today
as
it
is,
that's
all.
A
D
B
D
There'll
be
one
Ingress
Point
represented
in
a
specific
resource
and
what
I
mean
by
resource
is
like
a
subset
of
a
request:
trade,
a
subset
of
a
distributed
Trace,
and
that's
going
to
be
in
that
Ingress
point
of
the
service
boundary
yeah.
B
So
it
okay,
if
all
right,
I,
haven't
spent
a
whole.
This
is
all
gut,
so
I.
B
If,
if
we
like
per
Yuri's
comments
here,
like
there,
there
exists
these
interstitial
pieces
in
between
the
client
talking
to
the
server
rack
is
clearly
the
server,
because
that's
when
Ruby's
actually
handling
the
business
logic
of
responding
to
a
client.
But
some
things
in
between
that
has
caused
queue
time
and
we're
attempting
to
like
visualize.
That
ghost
insert
it
like
rack
knows
because
there's
a
header
that
would
tell
it
how
long
it
sat
in
the
in
a
queue.
B
B
Except
that,
if
I
were
to
instrument
my
back
end
with
istio
and
istio,
is
generating
spans
and
participating
in
the
trace.
My
clients
and
server
traces
are
already
not
adjacent
in
the
hierarchy.
We.
D
They're
coming
in
as
fans,
but
we
don't
know
if
they're
coming
in
Span
links
so
like
if
they,
if,
if
they're,
if
I,
don't
I,
don't
know
about
istio
instrumentation
right
now,
I'm
just
doing
anything
but
like
if
some
some
server
like
that
when
it
when
it
behaves
as
a
proxy
right,
it's
kind
of
like
think
of
it
as
a
message:
queue
right,
yeah
like
if
you're
instrumenting
Kafka
any
messages
that
comes
in.
D
Is
it
going
to
continue
the
context
propagation
from
the
from
what
the
inbounder
message
or
is
Kafka
gonna
treat
those
as
spam
links
for
its
own
internal
instrumentation
and
say
like
these?
Are
these
are
links
from
through
propagation?
So
why
wouldn't
the
proxy
implementers
do
something
like
that?
So.
B
A
B
It
a
span
for
a
proxies
handing
back
and
it
bounces
around
whatever's
happening
in
a
distributed
system,
and
then
the
proxy
is
the
one
that
returns.
So
it's
reasonable
to
visualize
that
as
a
span
that
the
proxy
handled
all
this
during
this
time,
at
least
that's
how
I've
seen
of
the
proxies
that
have
instrumentation
and
generate
Telemetry,
they
generally
they're.
D
B
B
Weirdness
man,
yeah
and
I,
wouldn't,
and
so
if
that
is
the
way
that
it's
going
to
be
I,
would
say
that
in
in
a
world
where,
whatever
is
proxying
in
front
of
your
rack
process,
isn't
instrumentable,
say:
you're
running
a
Heroku
and
they're
not
going
to
give
you
any
any
router
information.
You
know
this
is
best.
Effort
represent
the
time
it's
spent
in
heroku's,
router
or
whoever
your
provider's
router
is.
B
B
D
D
Like
I
know,
the
h8
proxy
sopa
thing
does
not
participate
in
a
transaction;
it
is
really,
it
is
out,
is
out
of
band
because
it's
processing
the
AJ
proxy
logs
and
just
basically
generating
traces
for
those.
So
it's
not
it's
not
even
representing
itself
in
any
way
and
there's
no
open
Telemetry
effort
going
on
for
migrating.
It
either.
B
Okay,
let's
talk
more
since
obviously,
I
have
opinions,
whether
they're
good
or
not,
we'll
figure
it
out,
but.
D
D
A
D
Java
Posse
was
like
an
old
like
like
the
like
one
of
the
biggest
podcasts.
Oh.
D
This
cat,
this
cat,
that
was
in
riot
games
and
then
went,
was
that
Netflix
and
then
went
to
Riot
I.
Don't
know
why
I
can't
remember
people's.
B
Names
we're
gonna,
go
hit
my
back
channel
here
and
see.
If
somebody
can
tell
me
about
a
thing:
I
won't
take
up
more
of
time
here,
but
yes,
Ariel,
that's
a
doozy
I
will
try
to
help
and
come
up
with
something
for
us
to
do
in
that
rock
band.
In
the
meantime,
this
PR
is
about
re-implementing
existing
behavior
in
our
instrumentation.
Is
that
right,
yes,
and
we
currently
generate
server
spans
in
the.
D
D
I'm
still
in
draft
mode,
so
I'm
trying
to
resolve
these
conversations
before
I'm
like
like
I'm
trying
to
like
have
a
whole
other
dietary
I've,
been
here
about
kind
of
like
the
configuration
and
at
one
point
it's
configured
and
all
this
other
stuff
so
like
basically
I'm
like
because
this
is
a
different
model
right.
Using
this
events,
yeah
API
is
a
different
model
than
using
the
middleware
model.
So
it's
like
I'm
trying
to
make
it
so
I
don't
break
anything
but
at
the
same
time
try
to
understand
the
existing
functionality.
D
But
I
would
like
to
get
this
so
the
reason
why
this
ended
up
happening
was
because
of
the
like
I'm
unable
to
currently
do
log
correlation
using
the
current
span,
because
we
write
our
logs
after.
A
Yeah
I
think
looking
at
sdos,
you
recommended
would
be
a
a
good
idea.
I
did
find
there
is
the
hotel
semantic
conventions
working
group
channel
in
the
cncs
slack
I
think
you
know
the
HTTP
semantic
conventions
group
is
definitely
active
there.
So
I
would
try
to
call
these
issues
to
their
attention
to
be
like
hey.
Are
these
things
that
you're
going
to
be
addressing
I?
Know,
that's
not
like
gonna
solve
it's
not
going
to
solve.
A
B
A
Yeah
and
I
don't
know
it
might
not
be
a
horrible
thing
if
your
eyes
can
stand
it
without
bleeding
to
look
at
the
the
Java
instrumentation
just
a
little
bit
more
closely,
because
I
know
that
stuff
was
it's
heavily
based
on
like
the
Java,
datadog
instrumentation,
so
I
know
this
stuff
is
based
kind
of
on
what
datadog
was
doing
for
Ruby,
so
if
they,
if
they
did
not
Port
that
over
I,
think
that
would
be
something
a
note
if
they
did
Port
it
over
it'd
be
interesting
to
look
at
how
they
did
it
in.
A
Know
if,
if
if
Shopify
is
using
this
stuff
at
all
and
and
Ed,
would
miss
it
so
I'm
not
sure
if
you
know
all
off
hand
Robert,
but
it.
A
A
Talking
about
the
right
now,
there's
kind
of
like
this
HTTP
proxy
span.
It's
a
fake
span
that
the
instrumentation
will
create
that
represents
kind
of
the
request
queuing
of
an
HTTP
request.
So
it's
kind
of
like
basically
there's
a
header.
Actually
Quest
start
like
it's
set
by
your
receiving
server
and
then,
when
rack
processes,
it
does
like
a
subtraction
between
type
that
time
and
it's
like
this
spent
x
amount
of
time
in.
C
C
C
No,
no
there's
it's
a
there's,
a
luo
one
that
he's
been
contributing
to.
B
C
Has
to
do
with
like
our
own.
The
way
we
deploy
nginx
I
think
is,
like
Lua
made
more
sense
and
we
were
working
heavily
with
the
team
that
owns
sprouting
at
Shopify
and
they
had
some
preferences.
So
it
was,
it
just
made
sense
to
do
what
they
wanted
us
to
do,
because
not
because
it's
like
good
or
bad,
but
just
because
like
effectively
like
we're
in
someone
else's
house
right
so
being
so.
Respectful
of
that.
C
Yeah
yeah
Francis
was
to
heavily
like
in
Guidance
with
that
one,
and
it
ended
up
pretty
good
like
we're
not
seeing
any
performance
implications
there
and
we're
actually
doing
some
Nifty
stuff
with
what
is
being
called
deferred
sampling,
where,
like
we,
won't
necessarily
create
an
nginx
span
unless
the
server
the
service
that
it's
connected
to
responds
back,
saying,
hey,
I,
actually
traced
myself
in
the
trace
response,
so
that
we
don't
have
just
like
these
floating
engine.
X
fans
when
the
like
Downstream
service
didn't
actually
keep
its
own
Trace.
C
The
the
the
pr
that
Ariel
put
up
Francis
shared
it
with
me
a
little
while
ago,
and
it's
inspiring
some
potential
good
work
for
one
of
our
bigger
applications.
We
haven't,
kicked
it
off
yet,
but,
like
I,
think
that
we
could
probably
do
some
really
good
things
using
that
approach,
we
could
probably
differ
and
kind
of
alleviate
a
lot
of
performance
pressure
we
might
be
creating
with
one
of
our
custom
sampling
strategies.
So
it's
really
really
appreciated.
It's
cool.
It's
exciting.
D
Awesome
I'm
glad
it's
helpful
I,
like
I
said,
like
I
just
got
questions
more
of
like
me,
trying
to
understand
like
what
functionality
we
should
keep,
what
functionality
we
should
ditch
and
I.
Think
Rob
and
I
are
going
to
try
to
figure
out
what
we
can
do
work.
You
know
reach
out
to
the
working
group
to
get
more
insights
from
them,
but
Rob
in
my
case,
at
GitHub.
We
we
use
proxy,
basically
as
our
as
our
proxy
and
all
of
our
spans
are
generated
from
the
AJ
proxy
logs
themselves.
D
Okay,
so
like
we'll
do
you
know?
We
have
functions
that
kind
of
just
like
fill
in
the
HTTP
headers
when
they're
absent,
and
then
we
generate
spans
and
they're,
not
semantically,
correct
right
right
because
we're
generating
them
as
server
spans,
a
single
span
to
represent
ta,
which
is
the
total
time
for
the
server
Downstream
right
and
that's
how
we're
doing
our
representation
there
now
istio's
coming
to
us
in
the
near
future,
just
not
today
and
I
haven't
looked
into
what
istio
does
so.
C
Honestly,
I
think
if
this
is
like
the
very
like
I
feel
like
the
person's
like
product
manager
had
it's
like.
If
you
feel
like
you,
can
push
this
quicker
without
having
to
worry
about
it
like
ditch
the
front
ends
fans
if
they're
getting
in
your
way.
For
now,
like
don't
I,
don't
know
it's
such
a
I
think
it's
such
a
cool
approach
that,
like
start
with,
like
your
your
use
case
of
what
your
MVP
looks
like,
if
you
don't
necessarily
eat
the
front,
ends
fans.
B
I
I
think
I'm
agreeing
with
Robert
that
if
we
at
least
bring
over
the
same
behavior,
if
we
do
make
ghost
front
end
spans
in
the
existing
Telemetry
having
re-implemented
it
using
rack
events
instead
of
middleware
do
the
same
behavior
and
we
have
an
open
issue
about.
Is
this
semantically
correct.
C
D
C
D
C
D
C
Yeah
I
would
say
if
it's
gonna
I'm,
like
just
for
the
sake
of
like
this,
is
a
cool
approach.
It's
interesting
it'd
be
nice
to
see
you
able
to
just
like
use
it
right
away,
I'd
like
to
maybe
take
her
on
it
with
one
of
the
apps
that
I'm
responsible
for
see
how
it
works
in
practice
and
see
if
we
can
use
it
elsewhere.
So,
like
I,
know,
I'm
not
going
to
use
front
ends.
C
Fans
I
know
that
we
shouldn't
necessarily
dictate
the
roadmap
of
this
stuff
or
the
feature
completeness
based
off
our
own
needs,
but,
like
we
kind
of
do
in
a
positive
way,
right
like
fake
Ruby
shops
will
influence
this
stuff,
and
hopefully
it's
for
the
benefit
of
the
community
but
yeah.
If
you
are
being
blocked
on
whether
or
not
to
add
this,
and
it's
adding
a
lot
of
complexity
Kick
the
Can
doesn't
mean
you
can't
either
add
it
later
right.
Yeah.
A
B
A
A
Be
extracted
into
like
a
separate
middleware
that
somebody
could
include
if
they
wanted
this
because
from
these
discussions
is
this
kind
of
sounds
to
me
like
this
is
going
in
the
future.
This
will
be
a
legacy
feature
because
I
think
Shopify
has
you
know
instrumented
nginx
directly.
Ultimately,
it's
like
you're
gonna
want
this.
You
will
hope
that
open,
Telemetry,
Finds
Its
way
into
your
intermediary
and
it
will
already
be
creating
a
span
and
past
be
passing
along
the
proper
space
context
for
you.
A
So
this
is
kind
of
like
a
stop
Gap
feature
until
people
are
living
in
that
world,
but
yeah
if
we
do
find
it
useful.
Maybe
this
can
just
be
like
a
middleware
that
you
can.
That
could
be
part
of
contrib
that
people
could
optionally
opt
into,
but
definitely
it
doesn't
need
to
be
something
like
it's
in
entangled
with
the
rack
events,
instrumentation
for
sure
yeah.
B
C
Because
this
this
event
handling
instrumentation
style,
this
different
approach,
it's
not
we're
not
intending
to
completely
throw
away
the
existing
middleware
approach
for
now
right,
it's
just
it's
a
fork
in
the
road
right
like
you
can
do
it
this
way
or
that
way,
and
hopefully
the
suspicion
that
this
is
better
because
it
looks
like
it
might
be,
just
better
that
eventually
just
becomes
the
default,
and
if
it
doesn't
that's
okay,
too,.
D
My
plan
for
the
release
is
add
another
configuration
or
not
even
right,
because
you
have
to
tell
people
well
I
guess
we
need
it
for
like
rails
and
for
Sinatra,
but
there'll
be
a
configuration
and
you
could
say,
like
you
know,
USE
events,
API
versus
you
know,
use
the
middleware,
whatever
the
the
version,
one
version,
whatever
I'm
gonna
figure
out
some
option,
name
eventually,
that'll
get
released.
We'll
do
another
release
where
it's
like.
Now.
Events
is
the
default
migrate.
D
All
your
stuff
over
that'll
be
a
incompatible
but
not
incompatible,
but
yeah
like
a
minor
version
but
whatever
to
Signal
people
and
then
eventually
like
we
can
drop
the
maintenance
of
the
second
of
the
middleware
one.
D
C
Just
a
second
well,
it's
just
in
my
head
and
we're
about
to
I
think
we're
hitting
the
End
of
Time
Ariel.
Thank
you
for
all
the
reviews
on
the
trilogy
stuff.
Oh.
D
I'm
trying
to
like
give
her
a
fast
like
feedback
so
that
we
can
get
this.
So
we
can
get
this
out
because
she's
like
doing
a
lot
of
great
work
there
that
I
don't
have
time
to
do
myself
like
no.
C
It's
you're,
it's
great
I'm,
just
literally
thanking
you
because
I'm,
like
all
the
feedback,
is
like
you're,
just
you're
covering
it
all
right,
I,
don't
have
any
experience
with
Trilogy.
My
assumption
was
like
we
could
just
more
or
less
try
to
do
parody
with
what
we
have
and
that's
kind
of
the
direction
that
I
gave
her
because
I
think
there's
a
team
internally,
that's
looking
of
using
Trilogy
to
try
it
out,
see
how
it
works.
So
yeah
they're
like
okay,
let's
add
Trilogy,
instrumentation
and
I
was
like
okay.
D
Yeah,
generally
speaking
like
how
how
we
use
it,
we
don't
do
any
so
we
never
leverage
the
features
like,
for
example,
like
the
the
existing
ones
that
we
have
for
exist
for
existing
instrumentations,
where
we're
adding
like
shared
attributes,
to
share
them
on
the
way
down
which
I
know
the
baggage
proposal
is
coming
up
in
a
little
bit
or
the
carrying
on
the
proposal
coming
up.
But
it's
like
that's.
D
Why
that
those
features
were
never
added,
because
we
don't
use
them
in
production,
but
they
are
interesting
features
and
it
might
be
very
interesting
to
use
in
the
in
the
active
record
span,
an
active
record
span,
processors,
where
it's
like
one.
Once
we
switch
over
to
using
the
the
not
spam
processors,
the
event,
notification,
processors
right.
C
The
downstream
spans
yeah
yeah-
that's
that's
exactly
I
was
thinking
too
it'd,
be
really
nice
because
we're
I,
don't
think
I,
don't
know
for
using
the
the
downstream
like
attribute
process
like
propagation
we're
using
it
for
HTTP
in
a
lot
of
places,
but
I,
don't
I,
don't
know
of
anywhere
for
my
sequel
right
now,
yeah
but
I
know
it'll,
be
helpful.
C
D
The
biggest
challenge
about
the
challenge
about
Trilogy
is
that
it
doesn't
maintain
its
configuration
state
after
it's
been
initialized
and
it
also
it
lazily
like
interprets
what
host
it
is
because
it
has
like
this.
Not
not
this
concept
of
like
there's
a
logical
host
that
it
connects
to,
but
then
it
actually
connects
to
a
database
host
because
it's
constantly
changing
its
database
connections
based
on
like
what
host
is
like
an
active
versus
a
passive
primary
and
stuff.
So
what
will
happen?
D
It's
like
you'll
see
there
that,
like
there's
like
a
connected
Helper
and
the
connected
host
helper,
is
actually
running
an
SQL
query
on
the
database
and
asks
it.
What
is
your
actual
host
name,
because
what
I
have
is
a
logical
host
name
and
like
you,
you
know,
and
the
injection
you
know
the
configuration
that's
injected
that
comes
in
from
rails.
Active
record
is
really
dense,
but
if
you're
not
using
active
record
and
you're
just
using
the
driver,
none
of
those
attributes
are
available
to
you.
D
So
you're,
like
I,
have
no
idea
what
logical
database
I
want
to
connect
to.
So
it's
it's
you'll,
see
that
in
the
code
where,
like
you'll,
see
that
in
a
Trilogy
instrumentation,
where
it's
like
I
have
like
some
funky
like
do
a
bad
thing,
I
think,
which
is
like
create,
like
a
local
instance
variable.
That's
like
underscore
Hotel
host
name
or
something
like
that
to
try
it
I
like
prevent,
like
because
you'll
fall
into
an
infinite
Loop,
where
itself
is
trying
to
instrument
trying
to
figure
out
what
the
host
name
is.
D
B
So
that
yeah,
my
thing
was
a
few
weeks
back,
I
had
asked
about
hey.
Is
there
any
prior
art
for
a
distribution
of
open,
Telemetry
Ruby?
If
a
vendor
say
or
a
company
wanted
to
wrap
it
and
make
their
own
distribution
and,
as
I
recall
the
feedback?
It
was.
Why
make
a
distribution
when
you
can
contribute
Upstream,
so
I
propose
the
things
that
are
different
about
honeycomb's
distribution
is
a
sampler,
some
config
that
we
don't
actually
need
and
a
baggage
span
processor
and
a
carry-on
as
discussed
in
the
past.
B
Where
so,
like
the
context,
scope,
attributes,
it's
I
would
like
all
this
stuff
in
process
or
I.
Would
like
all
this
stuff
on
all
child
spans
baggage
span.
Processor
means
a
thing
in
the
in
the
in
the
I
hesitate
to
say
the
word
context
in
the
context
of
honeycomb's
distribution.
Bag
of
spam.
Processor
has
a
specific
meaning,
but
I
don't
know
that
it's
a
great
name
as
an
upstream
concept,
so
I'm
proposing
that
I
open
a
PR
to
add
those
things
to
control
as
a
new
contrib
libraries
and.
B
B
Wouldn't
see
these
things
going
in
the
core
until
like
contact
that
oh
tap
about
context,
attributes
what
became
specked
and
then
the
carry-on
span
processor
would
probably
become
like
context
attribute,
processing
or
something.
D
Any
any
objections
from
the
other
members.
B
B
And
a
sampler
yeah,
the
sample
is
really
just
if
you've
given
if
you've
set
a
sample
rate
in
otel's
way,
add
it
as
an
attribute
to
span
so
that
honeycomb
knows
what
the
sample
rate
was,
since
that's
not
part
of
the
payload
of
Hotel
spans
these
days
until
the
sampling
Sig
figures
out
how
they're
going
to
communicate.
Where.
C
B
Sample
rate
yeah
yeah,
it's
an
attribute
upon
a
span.
We've.
C
Started
looking
at
using
the
consistent
probability
sampler
and
that
gets
written
into
the
Trace
State
yeah
better
just
a
little
bit.
It's.
B
It
is
not
something
that
we
want
it's
there
for
folks
who
are
like
mixing
and
matching
migrating
off
of
our
proprietary
instrumentation
and
onto
it's
a
way
to
say
this
is
the
sample
rate
and
if
you
say
it
the
same
way,
whether
you're,
using
RV
lines
or
using
a
hotel
it'll
get
you
it'll,
get
you
to
a
point
where
you're
still
saying
you're
doing
head
sampling
the
way
that
you've
been
doing
it
until,
like
Hotel
sampling,
settles
out
and
communicates.
B
B
Yeah
yeah
exactly
you
had
mentioned
that
earlier,
like
yep
yep,
it's
a
doozy,
so
that
one
would
be
like
a
vendor
specific
sampler
in
contrib,
and
we
would
absolutely
put
our
names
as
code
owners
for
it
and
not
expect.
C
B
C
Just
you
open
that
folder,
that's
like
vendor
names,
so
I
think
throwing
it
in
contributes
totally
reasonable.
For
my
person.
B
It's
a
it
is
a
testament
to
Hotel
Ruby
folks
that
the
the
some
of
much
of
the
impetus
of
making
the
vendor
distribution
was
to
smooth
the
configuration
and
usage
of
hotel
and
the
rooted
people
are
like
Upstream
hotels,
fine,
it's
Ruby,
it'll,
tell
Ruby
configuring
and
using
Hotel.
Ruby
is
simple
enough.
That.
B
D
A
A
B
A
My
two
sons
on
this
is
I
could
see
this
going
either
way.
To
be
honest,
like
I
feel
like
right
now,
our
contrib
is,
it
does
not
have
a
lot
of
vendor
stuff
in
it.
So
if
a
vendor
wants
to
own
something
in
there,
I
think
that's
fine,
I
think.
Alternatively,
it
would
be
possible
to
provide,
since
these
are
kind
of
opt-in
anyways.
It
would
be
possible
to
provide
these
as
additional
packages
that
live
outside
of
Ruby
could.
A
A
With
collector
contrib
is
that
that
whole
setup
is
a
mistake,
and
people
should
just
be
keeping
that
stuff
in
their
own
repos,
but
at
the
same
time,
I
think
you
know
there
is
precedent
there
that
vendor
stuff
goes
and
contribute.
So
if
we.
A
A
B
A
B
C
B
C
It's
like
this
is
only
for
honeycomb
and
honeycomb
alone.
I
think
I
have
completely
changed
my
tune
based
off
of
what
Matt
said.
So
if
it's
like
carry
on
like
send
us
your
carry-ons,
that's
fun,
because
it's
like
someone
might
have
a
use
case
for
that,
and
you
can
even
more
than
that,
you
can
just
show
people
what's
possible
right.
C
So
I
think
that's
interesting,
but
yeah.
If
it's
something
like
this
is
such
a
compatibility
with
honeycomb.
It's
like
maybe.
D
B
D
C
So
this
is
like
super
tangent
land
I'm,
not
gonna,
hold
anyone
hostage
here
but
like
that
is
something
that
I've
discussed
with
Francis
a
bunch
of
times
and
like
we
have
like
those
intern
dudes
of
kind
of
awesome
and
we're
talking
about
it
because
it's
like
carry
on
isn't
it's
almost
like
indexing,
all
your
spans
on
a
specific
attribute
and
we're
like.
Can
we
get
the
same
solution
without
doing
it?
C
And
it's
like
at
some
point
someone's
like
what,
if
we
stood
up
like
a
graph
database
next
to
our
staff
store
and
I'm
like
TV?
Maybe
right
like
there's
some
potentially
fun
experiments
in
our
future.
That
I
hope
we
get
to
pursue
because,
like
I
would
like
to,
like
you
think
of
like
Shopify
and
like
how
we
have
lots
of
tenants
and
one
of
the
things
that
people
want
to
do
is
like
slice
and
dice
the
data
based
off
of
like
a
particular
tenant.
C
If
they're,
you
know
maybe
trying
to
help
a
specific
Merchant
and
it's
like
well
I,
don't
really
want
to
write
our
tenant
ID
to
every
single
span
ever
I.
Don't
want
to
do
that.
But
that's
that's
carry-on
would
solve
that
and
we
wouldn't
have
to
talk
about
just
do
it
and
just
be
done
with
it.
So
I
don't
know
anyways.
It's
very
interesting
because
it's
like
can.
B
C
The
other
thing
that
I
think
I
think
interesting-
and
this
is
again
really
tangent-
is
something
I've
been
talking
about
with
few
people.
Different
people
is
like
having
your
data
store,
being
able
to
just
completely
flatten
your
traces
for
you,
you
get
like
how
much
time
is
spent
in,
like
my
sequel,
right
on
in
one's
van
right,
like
flatten.
All
your
like
summarize,
all
your
traces
I
would
love
to
have
a
tool
to
do
that.
D
B
C
Like
that,
I
have
Rags
fans
that
look
like
that
ever
expands.
That
say
like
total
time
in
dbe
total
time
and
whatever
other
historical
time
HTTP
and
like
I.
Have
these
like
these,
like
hacky
Roll-Ups
on,
like
my
rack
span,
so
I
can
say
like
let's
slice
and
dice
and
see
like
what
are
our
slowest
tenants
and.
B
All
right
now,
the
baggage
in
the
carry-on
I'll
have
something
out
see
soon.
D
You're
the
best
and
hey
I'm,
all
right,
I
hope
that
you
all
have
a
great
day
and
thank
you
for
your
help.
Oh
I
had
one
more
PR
that
I
threw
up
there.
If
y'all
could
take
a
look
at
it,
it's
for
a
option.
That's
in
the
existing
middleware
that
isn't
used
at
all,
but
the
code
is
there.
It
just
doesn't
do
anything.
Okay,.