►
From YouTube: 2022-04-22 meeting
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
It's
always
amusing
to
me
to
see
you
dressed
for
work
in
your
because
you
know
the
norms,
I'm
sure
for
japanese
workplaces
are
so
different
than
the
us.
C
B
D
Are
you
on
the
how
many
days
in
are
you.
A
I
got
hit
saturday,
so
it's
like
saturday,
sunday,
monday,
tuesday
wednesday.
It
was
like
day
six
and
I'm
basically
at
the
point
where
it
just
feels
like
I
I'm
getting
over
a
cold,
so
not
sure,
but
it
was
pretty
brutal.
The
first
couple
days
like
saturday,
I
was
late.
I
laid
late
in
bed.
I
couldn't,
I
wasn't
even
feeling
well
to
watch
tv
I
was
just
like
it
was.
It
was
just
feverish
and
bad
wow
ivan
kobe
yeah.
I
got
I
got
my
my
house
got
covered
hit
this
week.
A
A
Yeah
well,
my
wife
picked
up
probably
from
a
patient
since
she's
a
doctor
and
just
sees
so
many
people
yeah
by
the
time
we
knew
what
was
going
on.
I
had
it
and
one
of
my
daughters
have
it.
The
other
daughter
has
stayed
safe
so
far,
but
she
also
just
has
forced
herself
to
hide
in
her
room
all
the
time.
A
B
I
don't
think
jackie
uses
the
tool
chain
so
when
I
like
I've
been
using
java
18
just
to
run
the
build
and
it
works.
So
I
haven't
had
any
problems,
but
I
didn't
notice
that
constable
can
show
up
all
of
a
sudden
which
is
probably
related
to
that.
So
sometimes
cockpit
also
shows
some
weird
stuff.
So
there
are
these
cases
that
japanese
doesn't
understand.
Vegas.
B
A
This
is
one
of
those
things
that
would
be
great
if
we
could
get
the
the
great
and
mighty
trask
the
action
master,
to
figure
out
how
to
enable
our
bot
to
be
able
to
commit
these
things
to
to
pr.
E
A
B
F
B
A
F
A
B
G
B
Yeah,
I
I
knew
john
was
in
a
process
also,
so
I
thought
maybe
the
timing
will
overlap
a
little
bit
and
it
did
so
that
will
be
happening
over
the
next
couple
of
months.
I
think,
and
I
will
still
it'll
be
tetris,
so
it's
sort
of
managed
to
steal,
so
it'll
be
still
open
source
stuff.
I
don't
think
the
context
switch
in
total
is
that
bad,
but
I'll
probably
just
be
helping
with
hotel
java,
but
not
doing
anything
that
active
anymore.
B
B
A
No
problem
that
we
definitely
vendors
all
the
vendors
and
all
your
friends
like
find
out
if
those
people
who
are
really
interested
in
becoming
an
open
source
contributor
for
starters,
because
we
probably
need
more
contributors
and
approvers
as
well.
We
really
need
we.
If
there's
people,
if
they're
java
experts
we're
interested
in
helping
out.
It
would
be
very
helpful
so
fast.
A
You
work
with
a
lot
of
really
high
powered
java
experts,
or
at
least
in
the
same
company,
so.
B
F
F
James,
we
need
somebody
to
replace
somebody
in
the
aipac
time
zone
to
replace
honorag.
F
Oh,
so
that
we
can,
you
know,
kick
things
around
the
kick
things
around.
B
H
B
B
B
B
A
B
F
Yeah,
well,
I
do
have
somebody
we
have
somebody
starting
may
2nd
who's.
Quite
good
john
besuti
he's
out
of
france
he's
a
java
champion,
so
he's.
B
E
E
D
Too,
I
don't
I
I
don't
know
what
the
answer
will
be.
I
know.
There's
people
that
are
using
open,
telemetry
and
new
relic
has
is
a
java
shop,
so
maybe
someone
would
be
interested.
A
Yeah
get
char
on
the
case,
she'll
she'll
whip
somebody
into
shape,
that's
pretty
sure.
D
Yeah
yeah,
I
think
my
biggest
blind
spot
is
all
the
tracing
stuff.
Obviously
that's
a
big
piece
of
it
and
you
know
all
the
questions
that
come
up.
You
know
I'll
have
to
dig
dig
in
to
answer.
A
D
F
Yeah
and
I
was
going
to
say
you
know,
we
won't
leave
you
hanging
jack,
yeah,
myself
and
matteis
yeah
and.
B
D
I
A
I'll
probably
reduce
my
my
pr
like
the
number
of
pr's.
I
look
through
just
the
stuff
where
I
mentioned
at
least
initially
so
because,
as
a
maintainer,
I
still
end
up
get.
I
get
notified
about
everything
under
the
sun,
but
I
can
focus
on
the
things
I'm
actually
explicitly
mentioned
on
and
that'll.
That
will
help
focus.
D
I
think
I
think
that
you
know
another
blind
spot
that
I
have
is
like
context
and
history.
So
you
know
on
ra,
on
pr's
and
on
like
issues
that
come
up.
You
all
are
both
so
helpful
in
terms
of
like
establishing
no
we're
not
going
to
do
that
because
of
this
little
piece
of
history
or
this
pr
that
happened
a
while
back
and
yeah.
I
think
it's
it's!
It's
really
easy
for
me
to
misspeak
and
because,
like
I
just
wasn't
around
for
that
stuff.
F
F
To
like
queue
up
topics
for
this
meeting
to
kind
of
go
through.
B
We
should
have
been
using
the
rational
doc
more
actually
that
catch
or
find
or
whatever
that
was
I'd
forgotten
about
the
rational
doc,
which
john
found-
and
I
was
like.
Oh
that's
exactly
what
the
purpose
of
the
doc
was
just
completely
forgot
about
it,
but
yeah
we
can
going
forward.
We
should
document
some
of
these
more
tricky
decisions
better
in
that
doc,
probably
for
future
situations,
because
eventually
jack
will
also
move
on
and
it
just
happens.
So
we
just
need
to
keep
that
context
there
as
best
as
we
can
perfect.
G
I
A
A
A
C
D
Do
you
think
I
wonder
what
josh
sarret's
involvement
will
be
going
forward?
He
was,
he
was
reviewing
a
lot
like
all
the
matrix
prs
for
a
while
curious
and
what
his
his
kind
of
time
commitment
is
to
hotel.
H
A
My
feeling
kind
of
is
with
googlers
people
from
google
and
being
maintainers
that
google
doesn't
generally
let
people
stick
with
open
source
very
long
unless
they
can
really
justify
their
existence.
So
I
think
it'd
be
a
little
tough
to
have
a
maintainer
from
google
for
any
significant
amount
of
time.
G
J
D
J
D
J
Like
like
api,
or
something
like
that,.
D
E
J
Yeah
cool
that'd,
be
great
I'll.
Have
a
look
at
that
pr
as
well.
If
it's
still
open
yeah,
it
is
oh
cool
yeah.
D
And
there's
there's
something:
that's
there's
part
of
part
of
the
metrics.
That's
the
case
back
that
we
haven't
implemented.
Yet
is
this
ability
for
the
metric
reader
to
have
a
default
aggregation
for
each
instrument?
D
And
we
don't
do
that
yet,
but
that,
like
once
exponential
histograms,
are,
are
stable.
I
think
what
we're
I
think.
What's
the
follow-up
step
is
going
to
be
that
there's
going
to
be
an
environment
variable
that
allows
you
to
switch
histograms
to
use
exponential,
histogram
aggregation
for
if
you're,
if
you're
using
the
otlp
exporter
so
like
there
should
just
be
an
easy
way
to
flip
it
from
like
the
default
exponential
bucket
histogram
to
or
the
default
explicit
bucket
to
exponential,
because
there's
a
histogram.
J
A
D
And
and
so
that,
like
you
know,
the
idea
was
like
you
know,
an
sdk
could
choose
its
best
available.
Histogram
and
honor
originally
brought
up
that
like
no
that's
a
breaking
change,
that's
going
to
break
back
ends
potentially,
and
so
that's
that's.
No
longer
part
of
the
spec.
That's
been
deleted,
so
they'll
just
be
an
explicit
bucket,
histogram
aggregation
and
an
exponential
bucket
histogram
aggregation
and
an
easy
way.
D
E
D
Like
a
like
a
a
bigger
brush,
a
broader
brush,
so
you
should
be
able
to
say
I
I
think
at
your
exporter
level
that
this
exporter
prefers
exponential
histograms
for
for
all
of
its
histogram
instruments
and
then
that'll
just
switch
it
across
the
board.
F
F
D
F
J
Yeah,
oh
yeah,
I'll,
give
that
a
check
actually
awesome,
not
that
my
tick,
it's
just
a
great
tick,
but
you
know
well.
F
Yeah,
the
more
if
we
have
some
good,
what
we're
all
great
ticks
here.
I
hope
I
can't
help,
but
they
get.
F
E
E
J
Yes,
yeah:
that's
something
that
we'll
probably
use
some
places
in
atlassian,
so
awesome,
yeah,
that's
good!.
J
Or
the
spring
graph
coil
in
java,
isn't
that
the
java
instrumentation
okay?
Well,
that's
the
only
one
I
know
of
that's
pretty
recently
added.
I
think.
J
F
J
All
right,
but
there
is
a
there-
is
a
a
spring
graphql.
J
F
J
I
Spring,
do
we
know
if
spring
graphql
uses
graphql
java.
B
A
F
Yeah
onrock
so
planning
to
make
131
patch
release.
We've
got
two
things
in
there:
one
is
the
android
build
fix,
and
then
this
one
it's
kind
of
big.
So
that
was
my
only
thought.
B
F
It,
but
I
think
it's
we're.
B
Yeah
because
they're
pretty
confident
the
only
thing
I
like
sometimes
these
big
cherry
pick
fixes
tend
to
fix
one
small
case.
But
then
another
bug
comes
in
you
patch
again
and
then
you
end
up
with
a
ton
of
patch
coat
on
top
of
the
release.
So
that's
something
that
we'd
prefer
to
avoid.
If
there
isn't
enough
bake
time,
but
if
we're
relatively
confident,
then
it's
fine,
of
course.
So
that's
just
a
judgment
call
I'm
not
too
sure
what
it
is.
G
F
B
G
B
F
I
am
suspicious
that
something
is
something
else
is
the
matter
here,
because
it
seems.
B
B
B
B
F
B
F
The
clock
gets
on
the
span
yeah,
and
can
we
re?
Can
we
even
get
the
same
benefit
like
if
we
measure
it
ourselves?
Don't
we
then
lose
that
nanosecond
offset
benefit.
H
F
B
B
The
clock
stuff
doesn't
happen,
I
think-
or
at
least
it
doesn't
happen
correctly
yeah,
but
I
think
it
is
implementable,
like
I
don't
think
it's
a
impossible
problem
or
anything
like
that.
It's
just
not
implemented.
Maybe.
G
B
F
G
F
B
Even
like
it's
in
set
start
time
like
if
set
start
is
called
then,
instead
of
using
that
wall
clock
as
the
anchor
you
could
just
use
that
time,
because
it's
only
going
to
be
red
ones
anyways.
So,
even
though
it's
not
technically
a
clock
plate
that
would
be
sort
of
a
hacky
but
reasonable
implementation
in
this
like
it
still
requires
working
sdk,
but
it
shouldn't
be
that
much.
I
think.
F
Yeah,
so
the
metrics,
because
we
want
to
capture
start
and
we
we
want
to
record
the
metric
duration.
F
F
G
D
G
D
B
B
Sdk
already
supports
accepting
a
start
time
and
an
end
time.
It's
just
then,
when
you
create
another
sub
span
like
a
client
span
or
something
normally,
it
computes
the
time
for
that,
based
on
the
time
when
the
span
was
started,
but
I'm
pretty
sure
if
you
call
set
start
timestamp,
it's
not
going
to
be
based
on
that.
It's
going
to
be
based
on
the
sdk,
still
internal
clock,
that's
independent
of
that
time,
so
sub
spans
will
have
sort
of
slightly
and
again.
This
is
also
going
to
be
in
the
nanoseconds
or
something
in
practice.
F
So
if
set
start,
timestamp
is
called
on
the
it
would
have
to
be
so
okay
so,
but
then
this
is
a
case
where.
B
B
F
B
F
A
This
is
honestly,
I
mean
this
is
something
so
let
me
step
back.
A
second
open
telemetry
is
professing
to
kind
of
have
these
three
pillars
of
instrumentation
metrics
choices
and
logs
right,
and
the
thing
that
is
supposedly
making
us
special
is
that
we
have
this
integrated,
sdk
and
there's
all
sorts
of
stuff,
and
yet
we
don't
have
a
way
to
like
have
the
metrics,
instrumentation
and
tracing
instrumentation
interact
in
a
way
here,
so
that
the
data
can
be
properly
gathered.
A
This
feels
like
it's
something
that
the
this
really
needs
to
go
to
the
respect
at
some
level,
because
this
I
mean
that's
true,
I
mean
I'm
not
saying
I
don't
mean
I'm
not
trying
I'm
not
trying
to
be
super
pejorative,
but
if,
like
what
trash
says,
if
people
go
and
have
their
tracing
system,
like
sum
up,
the
total
span,
duration
and
their
metrics
and
they're
according
to
different
numbers,
I
mean
they're,
probably
not
going
to
be
different
by
a
huge
amount.
But
my
guess
is
it's
going
to
be
systematic
drift
right.
A
F
A
A
The
other
points
other
people
have
here
if
other
people
are
doing
manual,
instrumentation
that
interacts
with
that
at
this
they're
not
going
to
be
calling
that
start
start
time
stamp
and
therefore
their
span,
their
spam
is
going
to
be
potentially
up
like
the
stun.
The
start
end
is
not
going
to
line
up
properly
with
what
the
instrumentation
like
the
automatic
instrumentation
is
doing
right,
because
it's
going
to
be
maintaining
its
own
special
clock
to
do
with
start
time
stamp,
but
the
user.
H
A
Except
that
everyone
knows
this,
especially
in
java
system.current.
Millions
is
not
an
accurate
plot
for
any
state
right
like
it
gives
you
wall
time
and
then
that's
about
all.
It
will
give
you
right
and
everyone.
I
think
everyone
knows
that
wall
time
is
not
not
the
right
way
to
be
measuring
durations.
A
A
But
the
issue
is
that
I
mean
users
can
do
that
too,
but
if
there's
a
different
clock
and
they
have
a
different
system
current
time
noise
as
their
pin
for
where
they
start
like
the
numbers
are
still
going
to
be
awfully
weird.
G
B
I
suspect
that
most
languages
would
have
like
the
clock.
I
don't
think
is
in
the
spec
for
some
reason,
but
anyone
with
experience
with
tracing
would
have
done
something
similar
to
what
we
did
in
java.
I
think,
like
it's
pretty
standard,
to
use
nano
time
after
that,
first
wall
time,
it's
not
quite
as
standard
to
do
it
for
local
route
and
then
it
sub
spans,
but
I
think,
like
brave,
like
zipkin
clocks,
all
did
this.
I
think-
and
so
I've
seen
this
a
lot
increasing.
C
B
H
B
Yeah,
you
will
run
into
user
problems
if
you
don't
do
the
nano
time
stuff,
at
least
for
the
duration
yeah,
the
parent
to
subspan.
It's
not
as
important,
I
think,
but
even
then,
like
you
might
have
a
child
spend
before
the
root
span
because
of
some
time
change.
If
you
don't
do
the
parenting
thing,
and
so
someone
will
notice
it
in
their
traces
at
some
point
and
then
like
what's
going
on,
does.
A
B
B
C
A
B
D
I
suppose
another
way
to
do
it
would
be
to
add
a
method
to
the
span
interface
itself.
That
allowed
you
to
read
its
start
time
and
end
time
and
like
so,
you
know
whatever
start
time
you
associated
with
it
when
you're
building
it.
You
could
just
read
that
back
out
from
what
was
generated
from
the
clock.
B
We'd
for
this
precise
calculation,
we
need
the
nano
time
as
well,
so
that,
like
exposing
the
clock,
is
the
easiest
solution
for
that.
I
think
like
if
the
span
exposed
a
clock.
Okay,
that's
a
way
to
do
it,
because
we
need
both
of
those
not
just
one
of
the
times.
It
has
to
be
both
the
wall
and
then
animals
right,
which
is
what
our
clock
exposes.
F
I
have
a
question
about
the
single
clock
versus
clock
per
root
span.
Can
you
take
like
at
startup
if
you
take
the
current
time
and
you
take
the
you
just
figure
out
the
the
offset
the
nana
time.
F
No,
I
think
I'm
not
thinking
about
this
right.
I
was
trying
to
think
like
that.
The
nanotype,
the
difference
in
the
nanotime
from
the
like
the
nano
part,
the
micro
second
part,
and
then
always,
oh,
I
see,
but
you
can't
always
subtract
that
from.
A
I
mean
the
best
solution
I
think,
would
end
up
being.
You
do
have
some
auto
updating
clock,
like
I'm
always
suggesting
at
the
the
sdk
level,
and
then
that
was
just
passed
to
the
new
root
span,
and
so
it
was.
There
would
be
some,
maybe
not
an
auto,
didn't
clock,
but
a
atomically
updated
copy
of
the
clock
that
gets
refreshed.
A
F
B
F
D
So
then
we
don't
need
it
at
the
the
api,
the
top
level
api.
B
A
You
know
I
mean
the
way
to
in
the
in
the
interim.
In
the
meantime,
I
think
the
way
to
best
try
to
account
for
systematic
drift
is
to
make
sure
that
the
metric
you're,
the
internal
instrumentation
metric
clock
that
you
use,
however,
that
until
this
is
viable,
make
sure
that
it,
it
always
starts
right
before
and
ends
right
before
it
ends,
or
vice
versa.
Make
sure
that
you
you're
not
doing
like
the
start
and
then
metric
start
and
then
and.
B
B
Start
span
is
way
more
anymore
microseconds
than
in
spin.
I
think
so,
like
there
isn't
really
anything
we
can
do.
I
think
yeah.
Just
imagine
I
don't
have
a
goodie
like
that.
That
starts
spamming.
It
is
this
huge
thing.
Well
and
span
is
just
passed
to
the
blocking
queue.
So
I
would
imagine
that
starts
then
to
be
a
lot
slower
than
in
spam,
kim.
B
A
A
G
C
D
A
F
F
H
J
That's
yeah
we've
had
that
happen
before
sometimes
I
had
like
my
last
job.
There
was
like
a
security
group
that
would
block
queering
the
ntp
server
so
like
that
server
would
just
like
get
out
of
sync
yeah.
F
F
D
Yeah
that
sounds
good.
You
know
I
was
thinking
so
just
for
context
on
iraq
and
others.
D
So
I
want
to
add
two
two
examples
in
particular:
one
just
kind
of
a
basic
demonstration
of
installing
and
using
the
agent
and
then
using
the
micrometer
shim,
so
the
micrometer
shim,
I
think
the
best
kind
of
application
demonstrated
on
is
probably
like
spring
boot
with
this
the
actuator,
because
that's
kind
of
the
use
case
that
everybody
is
going
to
use
and
then
I
guess,
I'm
open
to
suggestions
about
like
best
kind
of
framework
to
demonstrate
the
agent
on
I'm
most
familiar
with
spring.
So
I
could
do
that
in
like
15
minutes.
D
G
B
F
D
A
F
D
F
Right,
we
need
to
use
the
library
instrumentation.
D
H
D
That's
nice,
then,
because
then
we
we,
we
can
still
we're
demonstrating
two
different
things:
we're
demonstrating
like
the
agent
installed
in
auto
instrumentation
and
then
library
instrumentation
from
the
you
know
open
technology
java
instrumentation.
So
those
are
kind
of
two
things
we
want
to
make
people
aware
of
yep
that'll
be
a
good
box
for
you.
F
A
B
D
A
A
H
B
D
Yeah,
so
you
know
there's
two
different
ways
to
do
that
we
could.
We
could
do
that
kind
of
which
is
library,
instrumentation
and
the
micrometer
shim
manually,
install
that
or
you
could
have
the
agent
take
care
of
everything.
J
A
Code
is
very,
very
confusing
and
I
think
it
has
some
serious
issues
with
context
propagation.
Still,
the
last
time
I
looked
at
it,
there
was
definitely
some
worrying
parts
in
the
context,
propagation,
immigration.
I
don't
think
you're
written
right.
F
A
A
I'm
not
sure
yeah,
I
haven't
looked
at
it
since
they've
moved
so,
but
I
know
the
last
time
I
looked
at
it
when
it
was
still
springfield
hotel.
There
was
still
there
were
definitely
some
context.
Propagation
strangenesses
and
a
lot
of
that
was
due
to
the
brave
api
which
it
was
based
on,
did
not
align
amazingly
well
with
the
hotel
apis.
It
was
a
little
tricky
to
get
all
of
the
great
apis.
F
To
line
up
and
be
implemented
correctly
yeah,
I
am
not
a
fan
of
bridging
tracing
systems.
I'm
going
to
try
to
say
we
don't
support.
You
know.
E
F
Do
support
bridging
metrics,
but
for
like
open
tracing
shim
and
the
sleuth
shim
it
just
yeah.
I
agree
with
you
john
it
just
it
feels
like
there's
so
many
dark
corners
and
things
that
it's
just
a
hard
hard
problem
and
it
doesn't.
I
don't
see
that
it's
that
big
of
a
benefit
like
just
pick
one
or
the
other
and
use
those.
F
F
B
J
F
J
It's
reasonable
to
fix
before
we
finish
up.
I've
got
a
quick
question
to
bring
to
you
guys,
working
with
the
instrumentation
sig,
trying
to
get
the
htp
semantic
convention
to
stable,
and
I
was
just
wondering
if
there's
anything,
we're
kind
of
reaching
out
to
maintain
at
the
moment,
and
I
was
wondering
if
there's
anything
that
you
think
is
missing
from
the
http
semantic
convention
or
something
that
needs
to
be
changed
before
we
go
ahead
and
raise
pr
to
mark
it's
stable.
F
Yeah
I
agree
with
this
is
a
big
one
to
me
and
I
I
fully
support
on
these
chain.
I
think
these
I
would
like
to
see
these
changes
go
in
for
stability.
J
Yeah
the
the
optional
verse
required
and
the
what's
the
one,
the
server
name.
F
Yes,
yes,
okay,
I
don't
care
so
much
about
the
server
name.
I
think
it
makes
sense,
because
the
proposed
changes
make
sense
to
me
on
the
server
name,
because
it
is
a
confusing
thing
and
our
instrumentation.
F
I
think
people
will
get
it
wrong
a
lot,
but
these
require
the
set
of
different
optional
things.
I
really
like
the
the
consistency
of
on
the
client
http
spans,
always
having
the
full
url
and
on
the
server
always
having
one
set
of
at
a
consistent
set
of
attributes.
J
So
those
those
open
pr's
right
now,
those
only
things,
there's
nothing,
there's
nothing
extra!
That
you'd
be
happy,
you'd,
be
happy
if
those
pr's
get
merged,
you'd
be
happy
for
it
to
be
stable.
F
Yeah
I
opened
one,
I
had
one
other,
but
I
can
live
either
way.
It
was
the
target
capturing
target
as
path
plus
query.
Instead
of
target
capturing
path,
plus
queries,
yeah.
E
F
J
J
Okay,
cool
well
feel
free
to
there
is
there.
Is
that
I
just
put
an
issue
in
the.
J
In
the
the
google
doc,
if
you
have
any
thoughts
there,
but
sound
yeah
just
feel
free
to.
If
you
want
to.
F
Oh,
I
have
a
question
for
you:
what
how
does
how
did
the
net
attributes
factor
into
http
semantic
convention,
stability,
effort.
J
That's
a
good
question
because
I
guess
there's
there
are
references
to
the
like
general
network
attributes
right
like
the
net
stuff,
so
I
guess
yeah
marking
them
as
markings
to
be
a
stable
would
require
the
net
wants
to
be
saved
or
at
least
require
those
ones
that
are
usually
not
to
change.
Yeah
yeah.
J
I
think
it's
kind
of
it's
a
good
point,
but
I
I
think
it
is
kind
of
assumed
that
those
won't
change
or
they
can't
change
once
once
the
hdp
is
marked
as
stable,
but
yeah
it's
a
good
point.
Maybe
it
should
probably
be
explicitly
called
out
somewhere
because.
F
F
E
F
J
F
Cool
sorry
to
take
it
over
time.
No,
no!
No
sorry!
We
we,
we
should
have
had
it
put
your
topic
at
the
beginning,.
J
F
All
right,
yeah
yeah,
thanks
for
reaching
out
and
yeah.
I
also
like
I
thought
that
this
was
a
good
approach
in
general
of
for
trying
to
get
stability
of
of
documenting.
I
thought.
F
B
F
Strategy
for
because,
if
you
can-
and
I
I
do
agree
with
getting
the
like,
if
you
can
get
the
maintainers
and
get
even
get
the
libraries
you
know
all
updated
to
follow
something
then
essentially
it's
the
de
facto
standard
and
then
also
it'll
be
easier
to
get
the
maintainers
to
put
their
check
mark
on
stability
for
whatever
that's
worth.
I
still
think
it
means
something,
but
I
just
haven't
been
hit
over
the
head
enough
times
yet.
F
Yeah
so
yeah
I've
talked
with
ludmila
a
lot
about
this,
and
so
a
lot
of
these
are
on
the
java
side.
We've
actually
kind
of
enforced
this
pattern
of
attributes,
so
server
attributes
getters.
F
So
we
we
don't
give
people
all
of
those
options.
We
only
give
them
server
name
target
and
then
on
the.
F
Client
side,
we
only
give
them
url
so
in
that,
in
this
way,
through
the
instrumentation
api,
we've
sort
of
already
narrowed
down
those
choices.
F
F
But
yeah
I
mean
some
of
there's
still
a
couple
changes,
but
we
would
be
happy
to
conform
and
we've
recently
split
out.
So
we
have
instrumentation
api.
And
now
we
have
this
instrumentation
api
semantic
convention
module,
and
so
this
is
kind
of
where
things
that
are
not
stable
live
so
that
we
can
stabilize
the
instrumentation
api.
E
F
J
F
The
thing
that
we
can't
the
thing
that
we
probably
can't
do
or
won't
do
for
a
long
time
or
that
will
be
really
hard-
is
to
retry
to
retry
modeling
or
the
redirect
redirect
and
retry
modeling,
because
from
an
auto
instrumentation
perspective
like
a
lot
of
times
our
instrumentation
of
say,
apache.
F
For
us
to
capture
two
different
spans
for
those,
the
original
and
the
redirect.
J
Yeah,
I
think
it's
just
that
the
retry
redirect
thing
it
just
has
to
be
whenever
possible,
like
just
realistically
instrumentation,
can't
always
go
that
deep.
I
guess
yeah,
but
no,
it
looks
like
that's
a
good
way
to
kind
of
it
looks
like
this
class
here
you
have
up
is
a
good
way
to
kind
of
make
the
instrumentation
pretty
consistent
across
http.
G
Cool,
well,
that's
great
yeah.
B
J
It's
slow,
isn't
it
it's
there's
always,
and
it's
always
slowed
down
by
my
new
details,
yeah,
but
it's
gonna
be
once
I
think
the
semantic
invention
in
general
being
marked
stable
is
like
to
me
one
of
the
most
important
things
in
hotel.
It's
just
so
much
that
can
be
done
once
the
once.
The
telemetry
model
is
just
consistent,
so
much
back
ends
can
do
with
that
data
kind
of
thing.
So
I
agree.
Yeah.
B
D
So
this
is
so
I
thought
I
thought
like
there
was.
There
was
a
lot
to
review.
Most
of
it
was
just
changing
signatures
to
add
the
additional,
the
additional
parameterization
and
in
in
what
wasn't
that
was
copy
and
pasting
the
code
from
like
the
double
reservoir
to
the
long
run
so
far,
so
that
that's
why
I
felt
okay
about
it.
I
was
going
to
take
a
pass
like
you
know,
assuming
this
assuming
we
felt
comfortable
merging,
I
was
going
to
take
a
pass
at
trying
to
simplify
some
stuff
before
the
release.
D
So
you
know
I
I
I'll
take
that
inevitably
take
a
much
closer
look
at
that
point,
nice,
but
I
agree
with
the
direction
and
then.
B
B
B
There
are
many
examples
of
this
sort
of
anti-pattern
of
subverting
the
java
type
system
by
returning
a
generic
here,
because
we
can
see
tu
are
here
and
then
also
just
used
as
the
return
type
right
away,
so
that
can
often
be
a
class
cast
exception.
So
I
saw
a
lot
of
this,
so
this
is
also
something
like
these
parameters
should
probably
move
to
the
aggregator
factory.
B
F
Honorary
was
your
concern,
the
public
apis
changes,
or
I.
D
So,
just
so,
just
in
a
nutshell,
this
means
that
our
data
model
better
reflects
the
the
the
protocol
buffers
right
because,
like
protocol
buffers,
exemplars
can
have
one
of
you
know
and
their
value
can
be
one
of
a
double
or
in
64.
B
Yep,
I
think
so
then
also
lungs
being
with
lungs.
That's
what
we
would
expect.
It
was
a
bit
weird
for
lungs
to
do
with
doubles.
I
think
like
at
the
back
end,
then
you
would
read
the
proto,
and
so
the
value
for
the
point
would
be
as
long
as
the
value
for
the
example
would
be
as
double
I'm
pretty
sure.
Most
veterans
would
be
quite
confused
by
that.
So
that's
sort
of
fixed
now,
with
this
pair.
D
F
B
D
Think
where
I
was
going
with
that,
and
so
like
you
know,
even
if
you
choose
to
have
a
long
histogram
instrument
when
it
gets
exported
over
otlp
the
histogram
data
points,
the
sum
would
be
another
thing
that
could
be
long
or
double,
but
the
proto
doesn't
allow
you
to
have
long
or
double
it's
always
double.
And
so,
even
if
you
have
a
long
instrument,
it's
some
is
always
a
double
right.
B
D
Yeah:
okay,
well
I'll
pay
attention
to
the
aggregator
factory
and
then
also
to
make
sure
that
long
histograms
have
long
exemplars.
B
That's
one
file
exactly
so
I
was
expecting
this
pr
to
be
a
few
minutes.
Just
like
I
thought,
moving
just
the
public
api
wouldn't
work
out
because
it
seemed
like
the
same
type
right
so
like
oh,
this
is
probably
just
going
to
compile
anyways
and
it
ended
up
being
an
insane
change
yeah.
So
just
like
that,
probably.
F
All
right,
well,
I
will
wish
me
luck
with
the
patch
workflow.
I
think
everything
like
it
keeps
getting
slightly
better.
I
think
now.
F
Feeling
pretty
good
about
that
yeah
all
right!
Thanks
for
joining
james
good,
to
see
you
all.