►
From YouTube: 2020-05-14 Go SIG
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
A
B
C
Yes,
yes,
he
was
a
great
intern.
He
left
for
big
table,
but
we
were
working
on
dump
truck
at
the
time.
B
And
now
he
so
now
he
works
on
our
monitoring
product.
He
joined
like
just
a
few
months
ago
he's
in
addition
to
tl
duties,
he's
also
wrangling
the
the
40
interns
that
we
have
so
he
met.
C
Oh
god
yeah
he
was
one
of
my
best
interns
ever
actually.
B
B
C
C
D
Ahead-
hi
christopher
hey-
I
just
I
just
wanted
to.
I
just
wanted
to
say
that
I'm
rotating
on
the
project
since
the
contract
has
finished,
I
have
to
have
some
open
loops,
like
especially
in
the
con
country,
represent
repository
where
I
have
added
the
request
for
the
database
instrumentation.
D
Maybe
I
will
find
some
time
to
finish
this
at
some
point
but
yeah
I
wouldn't
hold
my
breath
on
it,
yeah
so
yeah
playing
so
long,
and
I'm
thankful
for
the.
C
Fish
all
right
now,
thanks
thanks,
grasmere
you've
been
wonderful
and
appreciate
it.
I
understand
the
nature
of
paid
work,
so
thanks
very
much
and
do
let
us
know
if
we
can
do
anything
for
you.
C
C
D
C
D
All
right,
all
right
so
see
you
maybe
some
other
time.
Thanks,
rasmir,
good
luck,
there's
the
kinfolk
and
stuff.
C
Cool
all
right
so
we'll
take
care
of
that.
We
should
look
at
the
contributory
po.
So
let's
do
an
order
here.
Pr
review.
Let's
see,
I
was
gonna,
ask
morgan
if
he
wanted
to
speak.
He
usually
doesn't
come
to
this
meeting,
so
maybe
maybe
you'd
like
to
I.
B
I
was
just
popping
in
to
see
I
know
we
have
some
intern
hosts
of
interns
who
are
going
to
be
joining
in
the
future.
Who
will
work
on
go?
I
just
wanted
to
see
if
they
were
here.
If
they
were,
I
could
make
an
introduction,
but
the
interns
haven't
started
yet,
and
the
hosts
aren't
here
yet
so.
C
Well:
okay:
you're
welcome
to
continue
yeah
I'll
just.
C
So
we
released
0.5
yesterday,
thanks
tyler
thanks
everybody
thanks
anthony.
So
following
that
I
put
out
a
pr
last
night,
just
there's
been
like
code
moves
over
a
long
long
period
of
time
like
the
organization
in
the
metrics
directory
was
totally
chaotic,
and
this
all
I've
done
is
move
things
around
so
that
they
feel
right
to
me.
C
I
added
some
minor
comments,
but
it's
like
basically
a
no
up.
I
welcome
you
to
look
at
it
and
think
about
whether
this
code
layout
makes
sense,
but
it's
totally
not
urgent,
because
it
doesn't
change
anything,
but
I
was
thinking
that
this
will
be
a
precursor
to
adding
instruments,
because
if
you
were
to
go
at
the
instruments
right
now,
the
chaos
would
continue,
and
this
will
improve
the
next
pr.
C
Let's,
this
719
has
been
opened
and
discussed.
Well,
there
was
a
prior
pr
there's
some
history,
but
we're
making
progress.
I
have
not
reviewed
this
yet,
since
the
last
updates,
I
noticed
he
added
a
weight
group
and
that
was
going
to
require
me
to
think
carefully.
So
I
didn't
review
too
carefully
when
I
glanced
at
it
earlier.
A
Go
for
it.
I
noticed
that
we'll
get
to
a
later
pr
about
that
shows
the
example,
the
use
of
the
the
exporter
and
part
of
the
configuration
it
has
to
do
with
batching.
Okay,
I
had
not
noticed
that
configuration
before
as
a
user,
and
so
I
was
wondering
about
how
how
essential
that
is
to
using
an
exporter
like
this
to
configure
the
batching.
C
Understood
the
question
steve
there
are
options
on
the
batch
spam
processor
are
those
the
ones
you're
referring
to.
A
A
Yeah
npr
711,
which
is
whip
right
now.
It
does
show
configuring
the
the
scheduling
and
the
batching
and
stuff
like
that
and
okay.
C
Well,
the
I
think
the
high
level
here
is
that
there
are
two
processors:
there's
a
batch
processor
and
a
sort
of
synchronous,
processor
and
the
synchronous
processor
will
and
the
exporter
will
see
one
spread
at
a
time
and
the
batch
processor
will
see
multiple
spans
at
a
time
and
the
debate
in
702,
followed
by
seven
719,
has
to
do
with
the
desired
semantics.
C
So
that's
why
I,
but
I
got
involved
for
just
for
a
flaky
test,
and
I
don't
think
I
care
enough
about
this
to
to
debate
it
so,
and
I
think
I
think,
in
the
balance
of
like
reason,
I
think
he's
probably
right,
but
the
better.
The
better
behavior
is
for
what
he's
changing
it
to.
A
C
To
me,
the
answer,
I
don't
know
the
answer
in
the
metrics
world,
we've
I've
paid
more
attention
and
there
was
a
sort
of
motion
to
make
it
so
that
every
exporter
had
a
sort
of
simple
version
of
its
constructor
that
you
could
use.
So
it's
sort
of
like
just
install
this
and
then
it
would
choose
some
defaults
there
and,
of
course,
there's
lots
of
ways.
You
could
reconfigure
or
optional
behaviors,
but
there
should
be
a
default.
I
don't
know
what
the
default
for
jaeger
is,
for
example,
or
otlp,
is.
A
C
A
C
C
So
I
don't
really
know
what
the
default
is,
but
I
guess
there's
no
simple
setup.
So
there's
no
default.
Okay,
I'm
sympathetic
to
a
desire
to
fix
this.
I
don't
I
don't
have
an
issue
with
you,
but
please
feel
free.
F
To
yeah
look
looking
at
the
the
options
struck,
the
defaults
are
5,
000
milliseconds
for
the
schedule,
delay
so
every
five
seconds
and
then
512
for
the
batch
size.
So
it
looks
like
here
they're
just
trying
to
set
it
very,
very
short,
delay
and
a
small
batch
size
so
that,
as
you
run
it
you
see
results
very
quickly.
C
F
Oh,
what's
the
default
batch,
I
think
that
was
the
question
that
I
don't
know.
I
think
I've
always
specified
a
batcher
or
a
span
processor.
C
C
Okay,
thank
you.
Thank
you.
I
think
that
that
was
an
open,
a
good
question
and
I
I
have
been
mostly
focusing
on
metrics.
I
don't
have
an
answer.
Okay,
so
we
just
discussed
this
demo.
I
because
I've
only
been
focusing
on
metrics,
I
kind
of
if
I
see
a
change,
a
change
is
requested
here
and
it's
not
metrics.
I've
just
been
like.
Okay,
I'm
looking.
So
that's,
maybe
not
not
super
fair,
but
I
did
check
in
on
some
of
these.
E
To
be
resolved,
I
resolved
the
ones
that
were
resolvable,
and
then
I
there's
still
there's
still
some
issues
still
actually
here
yeah.
I
responded
well
I'll.
Try
to
keep
active
on
that
one.
I
need
some
help
and
then
on
the
demo
one
I
that
one's
pretty
it
looks
pretty
close.
E
I
had
asked
to
have
complete
kubernetes
configuration
for
the
deployment
and
it
looks
like
that
got
pushed
up,
which
is
the
intimidating
number
size
if
you're
looking
at
the
change
set
and
that
open
telemetry,
collector
demo,
but
the
actual
code
base
is
really
small,
so
if
you're
other
than
josh
and
not
working
on
the
metric
stuff,
definitely
worth
other
eyes,
it's
really
close.
The
change
requested
has
been
fixed.
It's
just
I
haven't
had
the
time
this
morning
to
get
back
at
it.
C
Cool
so
I'll,
when
I
see
motion
I
will
get
an
error
again
and
look,
and
the
rest
of
us
can
too.
C
G
Taylor,
I
don't
understand
your
last
question
last
remark
on
the
pull
request.
E
Sure,
maybe
we
can
kind
of
go
into
it
here.
E
It
is
so
the
the
question
that
I
was
asking
here
is
that
you
have
a
lot
of
you
have
expected
attributes
on
the
event
array
and
what
you
are
not
testing
is,
if
there's
a
one-to-one
mapping
what
you're
testing
is
if
the
produced
attributes
are
inside
the
expected
array,
however,
the
produced
attributes
sometimes
could
include
attributes
that
were
not
in
the
expected
set,
and
I
think
it's
a
better
idea
to
try
to
be
comprehensive
in
what
you're
actually
testing,
if
you're
going
to
test
against
the
expected
array
test.
E
All
of
them
in
that
sense,
what
you're
going
to
actually
catch
in
that
process
is
things
when
it
produces
things
that
you
weren't
expecting.
These
are
things
that
you
know
you
should
not
be
releasing
the
code
for,
because
that's
going
to
actually
cause.
D
E
Caused
the
the
user
to
have
additional
things
popping
up
when
before
they
were
getting
something
that
was
reduced
and
set.
Does
that
help
at
all.
C
E
Yeah,
I
suggested
that
as
well.
I
think
there
was
some
some
structural
differences
in
the
types
which,
if
you're,
if
you're
into
readjusting
those
structure
issues.
This
is
a
really
great
way
to
just
address
it
in
one
fellow
suit,
because
it's
it's
really
convenient,
so
yeah
I'd
still
recommend
it.
But
if
you
set
on
not
in
keeping
the
structure
of
those
types
different
than
yeah
as
long
as
we
check
the
comprehensiveness
of
that,
then
I
think
we're
on
the
same
path.
G
Yeah,
I
won't
go
with
the
structure
changes
because
I
don't
know
the
attributes
order
what
we
will
be.
That's
why
I
kept
and
kept
in
the
map,
but
I
will
match
the
expected
and
given
attributes
cool
awesome.
That
sounds
good.
C
C
So
overtell
spec
does
include
support
for
multi-valued
attributes.
A
That's
come
up
with
with
honeycomb
before
their
data
model.
You
can
pretty
much
submit
anything
that
can
be
serialized
in
json.
C
Yeah,
I'm
a
100
supporter
of
just
give
us
anything.
Json
I've
filed
issues
inspector,
though
about
nulls
being
a
valid
value
too,
and
I'm
I've
been
the
one
arguing
in
favor
of
map
value
and
list
valued
attributes.
So
I
totally
support
this.
A
The
one
thing
that
I'll
just
say,
though,
that
so
you
can
send
it
and
they'll
ingest
it,
but
when
it
comes
time
to
actually
use
it,
it
turns
out
to
be
rather
inconvenient,
like
often
sometimes
what
you
might
really
want
us.
Some
sort
of
set
semantics
against
it,
like
I
want
to
know,
does
this?
Does
this
field
contain
value
x?
You
know,
as
opposed
to
saying,
like
is
element
number
three,
this
value,
or
you
know,
checking
the
length
of
the
array
or
something
so
I
I
my
own
usage
of
it.
I
have.
C
Yeah,
I
I
hear
you:
it's
not
convenient
to
use
list
values
when
you're
going
to
be
querying
for
this
year.
D
C
Think
and
that's
another
reason
why
we
ought
to
go
as
far
as
having
map
values,
and
I
know
that
that
is
a
under
discussion
in
some
spec
issue.
I
don't
remember
the
number
but
yeah,
I
think.
C
If
we
added
map
values
and
null
values
list
values
are
already
sort
of
there,
then
we
have
all
the
values
the
json
value
set,
which
I
think
would
be
right
and
there's
some
improposition
and
they
know
where
it
comes
from,
like
you
have,
most
exporters
want
strings
for
your
attributes,
and
so,
if
you're
exporting
the
prometheus
like
and
you've
got
a
list
value,
what
do
you
do
and-
and
some
people
will
say-
that's
that's
impossible,
so
you
shouldn't
ever
allow
that,
but
I
think
I'll
just
put
that
with
a
string.
C
That's
not
a
big
deal
to
encode
my
json
thing
as
as
a
string,
it's
easy,
but
then,
when
you're
interpreting
it,
you
have
to
say
oh,
was
this
json
or
not
yeah,
because
I
have
to
parse
it
somehow.
Maybe-
and
you
know,
like
it
lights
up,
we
have
formats
where,
if
it's
json
we're
going
to
parse
it,
so
we
can
scrub
it
or
or
filter
it
somehow
yeah,
it's
super
complicated
because
you
have
to
have
two
different
string
fields,
or
at
least
do
some
inference
like
two
different
string
fields.
One
is
this
is
a.
D
C
So,
as
for
the
api
makes
sense,
I
believe
otlp
already
has
the
structure
for
it.
I'd
like
to
see
in
any
old
interface
that
can
be
jsonified
here,
but
I
don't
want
to
play
for
it
either
so
yeah.
We
should
do
this.
The
question
here
from
stefan.
I
think
the
answer
is
no.
A
A
C
Seems
like
I
used
to,
I
used
to
think
that
mattered.
A
lot
more.
I've
noticed
that
goku
has
gotten
really
good
at
optimizing.
Some
of
the
contemporaries
involved
in
variatics
so,
like
I
kind
of
don't
think
this
matters,
but
I
wonder
if
see
the
way
it's
been
specced
here,
it's
a
list
of
strings.
Why
don't
we
have
a
list
of
values
like
it
could
be
a
list
of
interfaces,
for
example,
that
would
be
better.
In
my
opinion,.
E
Yeah
to
keep
in
mind,
that's
that's
just
coming
straight
from
the
spec
because
remember
like
this
was,
I
think,
more
defined
by
bogdan
who's
working
in
the
java
stuff,
where
they
only
have
string
values.
So
I
I
yeah.
I
agree.
You
do
values
there.
A
A
C
That
you
can
add
that
dot,
okay
and
there
was
a
review
yesterday
where
tyler
asked
me.
If
I
had
done
this
intentionally,
it
was
like
there
were
three
cases
and
two
of
them
were
gonna
end
up
with
a
dot
dot
dot.
One
of
them
wasn't.
It
was
like.
I
don't
know
it's
sort
of
like
it's
sort
of
a
mixed
mixed
result.
C
C
I
I
still
miss
from
open
tracing
the
ability
to
have
a
lazy
value
type
where
you
can
put
in
a
struct
that
will
get
evaluated
into
a
json
struct
or
put
a
function
in
that
will
evaluate
to
something
to
an
interface
so
that
you
can
defer.
Instead
of
having
a
lazy
attribute,
you
have
a
lazy
value.
In
other
words-
and
there
have
been-
I
don't
even
know
where,
if
anything
is
left
in
the
hotel
spec
about
lazy
attributes.
C
C
A
As
a
user
of
such
a
feature,
I
would
want
a
commitment
about
when
it's
forced,
in
other
words,
in
the
face
of
like
sorry
just
talking
earlier
about
batching,
I
would
want
to
have
some
idea
of
say:
I'm
I'm
putting
in
a
closure
there,
and
so
I
believe
that
I
can
now
you
know,
unwind
my
stack
and
move
on
to
other
things,
but
I've
effectively
captured
a
closure
which
is
going
to
be
alive
for
some
amount
of
time
until
that
value
is
forced.
Yes,
I
have
an
opinion.
C
Here
too,
you
don't
want
to
let
this
deferment
happen
until
some
asynchronous
exporter
down
downstream,
because
you're
referencing
all
kinds
of
live
program
state
when
you
log
or
an
event
usually,
and
so
the
idea
is
if
the
span
is
actually
recording
evaluate
now,
if,
if
this
is
sampled
out,
just
don't
evaluate
it,
this
yeah
this
will
come
up
in
any
time.
We
talk
about
logging
api
as
well.
It's
dangerous
to
log,
a
pointer
to
something
that
you're
going
to
evaluate
some
random
point
in
time.
A
Yeah
because
it
introduces
concurrence
possible
concurrency
in
a
very
subtle
way,
if
you
exactly
about
when
it's
when
it's
forced
yeah,
it's.
C
Anyway?
We
are
way
off
this
ticket.
I
support
this,
but
I
kind
of
wish
we
could
move
the
spec
towards
having
this
b
interface
empty
interface.
Any
list
would
do
because
you
could
stringify
anything
essentially
and
when
we
ever
get
to
the
point
of
talking
about
recursive
structured
objects,
then.
F
Brought
up
with
the
spec
sig
meeting
last
week,
I
think
when
talking
about
the
the
log
spec,
whether
they've
got
that
any
value
data
type
in
the
log
definition
and
I
think
we're
going
to
want
to
reconcile
those
at
some
point.
C
Yeah
absolutely-
and
I
don't
know
if
anyone
out
here
has
ever
looked
at
recursive
structured
logging,
libraries,
because
there
aren't
very
many
of
them,
but
like
anyway,
I
should
stop
talking.
A
One
more
well
one
more
point
out
of
me:
anyway:
I
think
we're
being
distracted
a
little
bit
by
the
code
sample
there.
Reading
up
a
few
lines,
it
does
say
that
they
want
an
array
of
primitive
types,
so,
in
other
words,
there
would
be
as
well
an
ins
and
a
you
know
floats
or
whatever
other
types.
C
Who
cares?
Okay,
it's
all
gonna
get
stringified
at
the
end,
so
anyway,
this
is
a.
This
is
a
thing
where
I
don't
have
the
will
to
fight
the
whole
group,
so.
A
You
know,
and
just
to
give
a
supporting
case
in
one
of
honeycomb's
b
lines
that
handles
the
the
standard,
sql
library.
I
think
it
does
similar
for
others.
One
thing
it
does
is
it
captures
the
the
sequence
of
parameters
for
evaluating
a
given
statement,
which,
of
course
are
going
to
be
of
any
type.
A
You
know
it's
always
just
a
sequence
of
interface
and
go
so
it's
already
doing
this
and
I'm
pretty
sure
that
it's
not
just
stringifying
each
one,
but
rather
setting
them
as
the
closest
json
type
that
it
can
figure
out
in
that
scene
that
sound.
C
Makes
sense
to
me
I
mean
this
is
also
reminding
me
of
some
other
alarming
bits
of
the
spec
that
I
think
are
still
there,
which
has
to
do
with.
If
you
see
a
null
there's
something
about
you
just
drop
it
like.
Why
would
you
I
have
a
list
where
some
null
entries
are
there?
C
Exciting
here,
okay,
I
we
can
move
forward
on
this
issue.
We
should,
I
think
we
should
consider
pushing
a
bit
on
the
spec
and
making
this
interface,
not
a
string.
We've
already
got
this
infer
method
that
liz
contributed.
So
we
should,
if
you
took
an
array
of
interface
here,
you
just
would
infer
each
value
and
then
all
we
would
need
would
be
some
sort
of
new
value
support.
A
Let
me
ask
this:
could
you
could
you
still
comply
with
the
spirit
of
the
spec
and
implement
it
that
way
and
go?
I
think
so.
I.
F
C
Yeah,
I
also
think
that
we
have
a
lot
of
the
the
hotel,
spec
committee
or
the
technical
committee
kind
of
like
has
some
like.
They
can't
tell
us
what
to
do
and
go
and
and
by
the
way
bogdan
keeps
looking
at
the
metrics
api
and
saying
there
aren't
interfaces
here.
You
need
more
interface
and
I'm
like
nope.
I
don't
think
so.
This
is
a
go
discussion,
not
a
spectacular,
so
I
think
we
should
do
what's
write
and
go
and
worry
about.
There's
got
to
be
a
conversion
that
happens
because
of
the
spec
says.
C
The
protocol
does
this,
so
we
can
deal
with
that
in
the
conversion
and
we
should
just
do
what's
right,
so
yeah.
I
think
we
should
do
what's
right,
json
and
go
will
want
you
to
think
about
a
list
of
interface.
That's
the
natural
bariatic
type
of
list,
essentially.
C
Oh
here
we
are
sorry,
the
716
is
the
one
that
will
get
closed
by
my
reorg's
thing,
new
knee
drill
with
a
fair,
unified
meter.
This
was
tyler
suggested
a
nice
cleanup
that
someone
will
enjoy
as
a
first
issue.
C
This
one
708
is
asking
questions
about
how
to
do
prometheus
and
it's
essentially
blocked
on
the
upcoming
edition
of
new
instruments
or
approximately
blocked
on
the
upcoming
new
instruments,
as
well
as
some
exporter
work
and
some
design
work,
which
is
sort
of
like
what
I
expected
to
be
doing
for
the
next
two
weeks,
because
I
plan
to
get
us
to
a
point
where
we
have
an
sdk
spec
for
metrics
by
the
end
of
the
month
and
to
to
get
that
finished.
C
Changing
register
to
observer
was
a
was
a
recommendation
of
mine.
I
wanted
to
discuss
that
briefly
in
front
of
metric
sig.
I
don't
think
there's
a
meaningful
distinction
between
new
and
register
anymore
and
it's
confusing
to
me
as
I
maintain
the
code.
Why
did
I
choose
register
here
new
here.
C
The
drafts
back
pr
says
the
same
thing,
so
hopefully
that'll
get
accepted.
Okay,
finally,
one
I
don't
know
much
about
here,
we
go
okay.
This
is
the
issue
that
gets
fixed
by
the
open
pr.
C
C
We
there's
a
lot
of
people
asking
for
a
configurable,
metrics
sdk
right
now,
and
I
don't
know
how
much
time
I
have,
but
I
wanted
to
get
that
sort
of
experiment
with
that
and
it's
related
to
views
sort
of
a
question
about
how
do
you
configure
a
particular
output,
which
is
a
view,
so
I
think
this
will
will
make
progress
on
this
in
the
next
couple
weeks.
C
Add
a
test
grpc
trace
package.
That's
also
gotten
outstanding
pr,
as
I
understand
it.
So,
let's
see
this
guy
marara
keeps
asking
lots
of
questions.
It's
the
third
one,
sort
of
the
same
as
the
two
above
708
and
689.
F
F
I
think
some
sort
of
thing
that
would
get
cleaned
up
by
an
auto
closed
bot.
If
we
had
one
got
it.
A
Wasn't
this
josh?
I
thought
that
you
had
spoken
to
this
a
few
weeks
ago
when
I'll
probably
miss
quoting
you,
but
you
said
something
like
the
interface
is
right.
Only
you
know
you
that
you
don't
consume
telemetry
in
process
like.
D
F
A
C
I
actually
I
was
going
to
say
some
more
things.
I
I'm
aware
of
this
desire
and
it's
pretty
sensible,
it's
important
to
make
a
distinction.
I
think,
between
accessing
the
spam
and
accessing
the
command
context.
Currently,
the
hotel
spec
does
not
include
spam
id
as
a
spam
context,
property
nor.
D
C
A
spam
project
company
and
you
can
access
the
trace
id
and
the
span
id
of
a
spam
context,
because
those
are
sort
of
public
fields.
I
guess.
C
I
mean
I
I
you
know-
I
remember
just
have
memories
of
this
particular
topic
being
debated
in
the
old
days
of
google,
because
when
you
sample
out
a
spam-
and
you
don't
know
anything
about
it,
except
some
ids
that
were
decided
not
to
be
used,
it's
really
frustrating
and
I
kind
of
wish
you
knew
the
name.
C
So
I
would
almost
make
I
could
make
an
argument.
This
would
just
be
a
spec
level
argument,
though,
to
say
expand.
Context
may
copy
certain
essential
fields
from
its
span
make
them
accessible
like
the
name,
maybe
that's
it,
but
I,
but
that
would
be
a
a
pretty
significant
spec
change
experience.
Kennedy
is
probably
in
the
same
boat,
so
I
can
see
an
argument
for
it,
but
it
requires
a
lot
of.
A
A
F
F
C
D
C
You
added
a
version
of
the
library
itself.
This
is
about
the
provider
yeah.
Okay,
let's
put
this
in
the
next
milestone
or
yes,
great,
I
mean
the
same,
isn't
tracing
in
the
same
boat.
Do
we
have
version
optional.
E
Yeah,
it
definitely
needs
to
be
if
it's
not
already.
C
E
Yeah,
I
think,
I
think,
with
your
recent
changes
to
the
spec
api
you've
included
a
more
descriptive,
batching
or
timestamp
that's
associated
with
it.
It
was
kind
of
nebulous
whether
the
timestamp
should
be
per
record
or
whether
it
should
be
per
batch.
I
think
you
laid
out
better
guidelines
on
it,
so
this
actually
could
probably
be
addressed
at
this
point.
C
Yes,
I
think
that
there's
something
should
be
done.
The
spec
says
that
the
timestamp
is
implicit
so
that
the
sdk
can
make
a
decision
about
what
it
wants,
but
it
also
says
that
all
the
observers
are
snatched
out
in
one
instance.
So
I
think
we
do
need
to
have
a
single
time.
Stamp
has
to
be
provided
for
the
interval
and
actually
have
some
relevant
discussion
about.
C
F
It
does
bring
up
one
thing,
though,
which
is
when
I
was
adding
the
trace
state
propagation
a
bit
back.
I
found
that
the
w3c
pre-state
as
potentially
multiple
headers,
that
should
be
combined
if
possible,
but
our
current
propagator
interface.
Only
the
supplier
only
gives
back
a
single
value.
You
can't
get
multiple
values
for
a
header,
and
I
I
think
also
it.
It
assumes
that
it's
text
based
which
would
complicate
the
the
binary
propagator
if
we
ever
tried
to
reimplement
that
that's
another
reason
why.
C
Like
I
think,
there's
so
little
value
to
the
world
and
having
a
binary
complicator,
let's
drop
it,
but
that's
not
my
decision,
but
but
kafka
headers
are
bite
slices.
I
never
forgot
about
how
to
turn
a
string
into
a
bite
or
a
bite
slice.
Okay,
this
one
just
it's
just
like
so
much
work
for
so
little
value.
That's
all!
I
have.
C
The
thing
about
headers,
though
I
thought
I
recall
some
vague
corner
of
my
memory,
about
the
go
as
the
go
library
itself
treating
multi-valued
headers,
oddly,
which
sort
of
falls
through
to
this.
I
don't
remember
the
detail.
It
was
one
of
my
former
colleagues
that
uncovered
it
and
documented
it
for
us
it
affects
grpc
and
various
things,
so
there
could
be
some
problem
there
in
the
like,
the
literally
the
net
http.
C
F
Yeah,
so
this
one
it
looks
like
they,
they
were,
assuming
that
the
this
bundler
was
used
to
count
bytes,
where
it's
in
instead,
I
think
just
being
used
to
count
items
that
are
being
put
in
there,
and
so
it's
assuming
that
each
item
is
one
bite.
F
It's
it's
a
bit
weird
use
of
the
bundler,
but
I
don't
think
it's
wrong.
What
was
a
bundler
is
that.
F
C
The
library
that
did
this
in
the
world
we
have
one
internally,
it
sucks
wow
god
is
the
worst
library
to
write
and
go
because
god
doesn't
have
any
generics.
F
C
C
F
H
C
Okay,
so
is
the
action
item
here
to.
E
Yeah,
I
think
we
could
just
close
it.
Then,
if
that's
the
case,
I
would
trust
you
on
that.
Maybe
just
leaving
a
comment
along
the
lines
of
reopen.
If
this
isn't
what
you
were
expecting.
C
C
Library
info
is
probably
the
same
as
library
version
are.
These
duplicates.
E
I
don't
think
so.
I
was
just
looking
at
this.
I
think
this
also
has
to
do
with
implementation
and
use
in
the
otlp
exporter,
including
it
in
a
resource.
Oh
and
that's,
that's
under
discretion
in
the.
C
E
Repo
yeah
yeah
so
unresolved
as
well.
E
I
might
actually
I'm
gonna
remove
this
from
the
next
milestone
and
then
maybe
put
a
label
on
this
thing.
C
I
actually
I
commented
on
this
pr
yesterday.
It's
just
like
no
good
answer.
I
think,
unfortunately,.
C
If
we
treat
these
as
metric
attributes,
I
think
it's
going
to
get
very
confusing
for
the
exporters.
If
we
treat
these
as
resources,
I
think
it's
correct,
but
it
doesn't
seem
to
like
it
because
then,
then
you
end
up
with
a
resource.
You
end
up
with
exporting
many
resources
per
batch
and
which
I
think
is
fine,
but
for
legacy
consumers
it's
pretty
tricky
to
deal
with.
I
don't
know
even
lightstep
doesn't
like
that.
C
So
I
think
the
ideal
outcome
would
be
a
better
protocol
that
can
deal
with
this
type
of.
E
It
yeah
I'm
pretty
sure
that
actually
already
works
well.
At
least
okay.
I'll.
Take
a
look
at
this
okay.
C
Well,
we're
almost
like
a
month
old
issue
and
we're
we're
going
to
run
out
of
time
to
talk
about
them.
I
mean
we
should
just.
E
Yeah,
I
think
that
the
only
other
thing
I
added
was
it
looks
like
the
v04
spec
release
closed
two
days
ago,
so
we
should
probably
try
to
make
sure
our
next
release
is
targeting
that
compatibility
and
implementation.
C
E
Take
on
the
matrix
updates,
okay,
cool
yeah.
I
took
an
action
item
last
time
that
I
didn't
follow
up
on
to
make
sure
that
we
have
all
issues
covering
all
the
topics.
I
I
know
that
there's
one
issue
that
isn't
created
and
that's
just
implementing
all
the
instruments,
but
I
just
figured
josh-
should
make
a
pr
without
even
an
instrument.
E
E
Yeah,
and
so
I
think
we
should
I'll
try
to
find
some
time
over
the
next
week
to
make
this
all
more
project
managed.
I
guess
at
this
point
one
of
my
weaknesses
for
sure,
but
I'll
give
it
a
go.
C
So
this
is
the
change
that
president
referred
to.
I
asked
someone
at
lightstep
to
look
at
this
because
it's
huge-
and
I
just
inspect
so
great.
E
Yeah,
the
big
takeaway
that
I
looked,
I
briefly
looked
at
it
it's
it.
There's
like,
I
think,
like
an
11
000
line
file
in
there
that
was
generated
yeah
and
the
generation
code
is
something
that
kesmir
had
like
built
as
well.
So
it's
like
relying
on
one
of
his
projects
like
I
tried
to
not
look
at
that
file
and
it
was
still
a
massive
pr.
So.
A
This
is,
this
is
the
same
thing
I
was
describing
earlier,
where
honeycomb
has
some
libraries
for
this
that
are
way
smaller.
I
don't
know
why
I
there's
my
handwritten
what
it's
worth.
C
E
I
think
the
important
thing
on
this
one,
though,
is
that,
as
customer
said
at
the
top
of
the
hour,
the
there's
no
owner
of
spr
anymore.
So
if
there
is
somebody
on
the
call
who
is
interested
in
wrangling
this
or
trying
to
pick
it
up,
leave
a
comment
or
hit
us
in
twitter,
and
we
can
we
can
work
on
making
that
happen.
E
I
think
there
was
also
some
movement
in
this
repo
josh
on
the
related
pr's
of
the
data
dog
headers.
It
sounds
like
there
was
a
comment.
I
think
in
15
that
someone
from
datadog
explicitly
said
like
yeah,
you
don't
need
to
include
attribution
to
datadog.
It
can
just
it
can
be
unified
as
the
open
telemetry
authors.
So
you
just
need
to
shore
that
up-
and
this
should
be
unblocked
at
this
point
it
sounds
like
so
yeah,
okay,.
E
It
needs
to
get
all
the
attribution
cleaned
up,
but
yeah
after
that.
I
think
it's
ready
to
go.
E
Josh,
I
don't
know
what
happened
your
microphone,
you're,
extremely
quiet,
all
of
a
sudden,
but
am
I
still
so
quiet?
It's
better
you're
still
under
an
ocean,
but
it's
it's.
I
don't
know
if
it's
coming
from
your
computer
instead
of
your
headset
or
something
like
that,
but
I
can
hear
you
now:
okay,
oh
there,
it
is
okay,
that's
way
better.
C
Okay,
my
my
computer
has
weird
audio
issues.
I
just
opened
my
laptop,
that's
all
I
did
and
I'm
still
on
the
thunderbolt
screen
up
here.
So
that's
weird,
okay
and
then
this
macro
and
plug-in
also
lost
a
maintainer
michael
michael
smith,
left
lightstep.
So
unfortunately
we
have
two
approvals.
E
It
sounds
like
there
is
a
ask,
because
the
span
name
is
going
to
be
based
on
the
path
and
currently
the
path
doesn't
have
some
sort
of
abstraction
for
high
cardinality
entries
like
username,
or
something
like
that,
which
other
things
have
and
kasmir
had
offered
some
suggestions
on
how
to
approach
that
one
being
kind
of
a
lazy
and
function
kind
of
like
what
you're
talking
about
before
or
the
other
being
to
try
to
submit
to
the
upstream
project
to
have
them
support
that.
E
So
the
lazy
function,
I
think,
was
the
way
that
mike
was
talking
about
updating
this
but
yeah.
He.
He
also
responded
somewhat
recently.
So
I
don't
know
if
he's
like
very
minimal
amount
of
time,.
C
Yeah
he's,
I
believe,
he's
looking
for
work
and
is
continuing
to
be
involved.
It's
an
unfortunate
situation,
but
he's
not
bitter,
so
he
seems
to
be
useful
at
this
point.
We're
hoping
he's
we're
hoping
to
find
him
hotel
work
because
there's
a
lot
of
people
looking
for
a.net
hotel
person,
he
was
primary
in
into.net,
not
in
go
so
okay.
I
feel
like
I'm
talking
about
things.
I
don't
really
follow,
so
I
should,
but
I
should
stop,
but
okay,
that
covers
20,
15
and
32.
C
that
one
was
blocked
on
the
same.
Did
it
ryan
byrne?
Here
is
the
one
I
was
referring
to
he's
been
doing
rc
plus
plus
work
and
is
incredibly
good.
So
I
I
he
but
he's
pretty
new
and
go
so
I
I
I'm
gonna,
ask
him
to
look
at
32
actually.
C
F
It's
adding
tracing
for
an
orm
system,
oh
yeah,
of
course,.
F
Yeah-
and
I
know
rahul
had
asked
for
some
changes
to
be
made
a
while
back
when
this
was
submitted,
and
I
don't
think
there's
been
any
activity
on
it.
Since
okay.
E
Yeah
march
5th,
so
this
might
end
up
being
stale
at
this
point,
should
we
close
things
that
are
stale
like
it
looks
like
useful.
E
I
don't
have
an
answer
to
that
question.
I
also
wonder
if
this
could
be
supported
using
the
new
database
thing,
but
I
that's
a
wild
guess
at
this
point.
C
Wondering
if
this
would
a
sql
driver,
instrumentation
package
be
suitable
tracing
form?
That's
my
question.
C
E
Yeah,
the
the
datadog
stuff-
I
I
could
probably
jump
in
there
because
I'm
pretty
sure
that's
going
to
require-
maybe
some
elevated
permissions
I'll
see
if
I
can
clean
that
up,
because
if
I
remember
correctly
outside
of
the
attribution
there
wasn't
any
issues
with
these
prs
so
I'll
try
to
get
that
done
tomorrow,
probably.
E
C
C
Well,
everybody
great
meeting
there's
a
few
minutes
left,
but
I
always
want
to
take
a
break
between
this
meeting
and
next.
So
I'm
gonna
drop
off.
It's
been
lovely,
keep
up
the
good
work
and
I'll
see
you
next
time.
Thanks
all.