►
From YouTube: 2020-12-04 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
Hello
again,
hey
I've
been
writing
benchmarks
for
baggage,
with
all
that
w3c
validation
in
the
baggage.
Putting
things
into
baggage
is
super
expensive.
A
A
A
I'm
not
sure
anyway,
something
we
can
something
we
can
think
about,
but
yeah
right
now
the
baggage
like
adding
items
to
baggage
is
pretty
is
quite
expensive,
like
adding
100
items
average
is
like
500
milliseconds
like
a
lot.
That's
pretty
ridiculous!
A
lot
yeah!
It's
it's
remarkably
slow!
Well,
I
started.
I
started
writing
benchmarks.
I
was
like
oh
let's,
let's
see
how
the
like
parenting
is
for
the
context
and
all
that,
like.
B
A
A
B
A
C
A
Yuri
for
help
yeah
yeah-
I
I
didn't
have
any.
Let
me
catch
you
up
a
little
bit
on
what's
going
on
via
about
the
release
release
meeting,
it
definitely
is
sounding
very,
very
likely
that
we'll
need
to
rip
all
the
metrics
out
of
the
main,
open,
telemetry
apis
before
release
and
move
all
that
into
some
experimental
module.
A
I
do
I
mean,
but
I
wonder
what
this
means
for
like
we're
using
metrics
in
the
sdk
like
what
should
we
do
with
that?
I
don't
know
like
we
use.
We
use
it
to
measure
the
q
depth
of
the
span
processor
and
over
the
batch
span.
Processor
and
the
number
of
items
we
drop
from
the
spam
processor,
like
is
our
sdk
allowed
to
use
experimental
apis?
A
Anyway,
I
mean
I
understand,
the
whole
point
is
if
we,
if
we're
actually
obeying
semver
and
we
release
something-
that's
called
1.0,
and
then
we
proceed
to
break
all
the
metric
apis
repeatedly
like
do.
We
need
a
major
version
for
every
time
we
break
a
metric
api
people
will
probably
just
get
super
annoyed
with
that.
C
C
A
Yeah
trask-
and
I
would
I
just
mentioned
on
the
meeting
I
don't
know
if
this
morning
or
whenever
it
was
I've,
been
so
many
meetings
like
do
we
need
to
remove
the
metric
exporter
support
from
the
agent?
B
Yeah,
I
think,
on
our
side.
We
have
a
lot
more
flexibility
because
of
no
at
least
where
we
don't
have
a
public
api.
I
don't.
B
A
B
Mean
I
think
my
interpretation
is
that
he
they
got
the
message
that
we're
not
ready
what
you
know.
What
can
you
do
if
we're
not
ready.
A
B
A
A
C
A
A
A
Open
telemetry
global
we'll
need
to
remove
metrics
from
that
as
well,
and
the
sdk
global
or
the
big
sdk
class.
That
part
will
be.
I
mean
mostly
deleting
code
right,
but
it's
still
no,
it's
work.
Of
course
you
got
it
yeah.
I
mean
it's
just
it's
all.
Just
it's
just
work.
It's
not
like
it's
incredibly
difficult,
but
it
will
slow
things.
It
will
take
some
time.
B
A
B
A
B
B
B
John
nix,
that,
with
the
the
compelling
argument
of
how
do.
C
C
C
Okay,
nowadays
like
somewhat
conventional,
to
use
an
annotation
to
declare
unstable
apis.
So
you
could
just
add
that
to
every
single
metric
api
and
if
you're
using
this,
you
have
to
worry
about
it
breaking
I
don't.
I
don't
think
I
would
use
it
if
we
have
a
very
high
chance
of
breaking
stuff,
it's
much
better
if
it's
like
this
is
probably
how
it
should
be,
even
though
it
might
break
then
we'd
use
an
annotation.
I
think
I'm
not
sure
where.
A
I
still
have
a
I
have
a
hunch.
I
have
no
evidence.
No
one
suggested
this
yet,
but
with
openmetrics
applying
to
be
a
member
of
the
cncf
and
putting
in
the
making
op
they
put
in
a
proposal
to
some
standards
body,
I
don't
remember
which
one
to
make
openmetrics
the
standard
metric
exposition
format
that
people
will
start
demanding
that
open,
telemetry,
metrics
apis,
look
more
like
open
metrics,
in
which
case
we'll,
basically
throw
all
of
our
apis
metrics
apis
away
and
start
from
scratch,
which
nobody
is.
A
C
A
C
B
C
A
A
Too,
anyway,
no
official
decisions
being
made,
but
ted's
writing
the
otep
right
now
and
I
think
it's
going
to
be
submitted,
maybe
in
the
hopefully
in
the
morning,
and
then
it's
more
open
for
public.
I
mean
the
document's
open
for
public
comment
now,
but
I
think
he's
going
to
try
to
codify
it
as
no
tip
so
we'll
see
what
happens
over
the
next
few
days.
B
C
B
C
B
We
did
do
the
zoom
dance
of
everybody,
log
off
and
log
back
in,
to
create
a
new
zoom
recording
so
hopefully
not.
A
C
B
So
a
bunch
of
discussion
about
nested
client
spans,
so
someone
was
kind
of
getting
bugged
and
sort
of
up
to
speed
on
some
of
our
use
cases
for
where
this
comes,
because
at
first
his
first
thought
was
well
the
the
higher
level
one
should
be
an
internal
span
and
and
then
only
the
leaf,
one
should
be
a
client
span,
and
so
we
kind
of
talked
through
the
example
of
both
what
you
wanted
from
the
aws
sdk,
the
client
modeling,
how
we
get
richer
info
at
the
higher
level
about
where
we're
connecting
to
like
s3,
so
that
and
and
sort
of
at
the
end
of
the
day
it
boils
down
to
for
all
of
the
vendors.
B
We
want
to
draw
that
up.
That
map
just
distributed
tracing
map
in
the
back
end,
and
so
that's
why
we
want
the
client
span
at
the
higher
level
span.
B
Okay,
there
was
also
the
yeah
so
where
to
do
context,
propagation
bogdan
was
thinking
so
yeah
context.
Propagation
would
happen
wherever
the
client
span
is
defined,
which
is
what
we
do,
which
makes
sense,
which
also
maps.
Well.
I
was
explaining
the
the
aws
example:
how
actually
it
we
have
to
do
context
propagation
at
that
layer?
Otherwise,
it's
signed
and
we
break
the
signature.
B
I
think
there's
still
some
question
about
like
this
database
client
like
in
the
elastic
example.
I
don't
know
if
you
can
put
headers
on
that,
but
presumably,
if
their
back
end
cared
about
distributed,
tracing
headers,
their
api
would
need
to
have
some
way
to
set
those.
B
So
I
think
that's
fine
nikita
was
a
little
concerned
about
that.
So
I'll
try
to
remember
that
in
case
that
comes
up
again.
C
B
Yeah,
this
was
a
good
example
that
convinced
blogden,
that
we
really
like,
we
don't
even
know
like
we
can't
even
make
that
decision
of
whether
it
should
be
an
internal,
whether
there's
something
underneath
it
at
that
time.
So
we
can't
flip
between
internal
and
clients,
so
it's
better
for
it
to
be
client
and
suppress
or
capture
that
as
events,
the
more
low-level
details.
B
I
was
just
writing
notes,
as
people
were
talking
so
bogdan,
didn't
bogdan
was
didn't
think
they
should
be
client.
He
thought
that
they
should
be
internal
spans
and
only
the
network
calls
should
be
on
clients
pans,
but
then
this
example
sort
of.
B
Yeah,
so
that
was
okay,
nested
client
spans.
I
think
that
was
everything
client
span
in
context.
B
Oh
yes,
yes,
this
was
interesting
discussion
about
how
we
put
the
client
span
in
the
context
to
know
if
we're
nested
and
bogdan
did
not
like
that
as
a
sort
of
a
hack
of
of
the
model
of
the
this,
the
modeling,
and
so
he
he
proposed
something
that
actually,
I
think,
is
quite
reasonable,
which
in
our
case
since
we
can-
especially
in
our
case
since
we
control
the
the
implementation
we
could
cast
to
the
problem,
is
we
can't
check
if
the
parent
is
what
the
kind
of
the
parent
is,
but
we
could
expose,
we
could
cast
in
our
instrumentation
or
we
could
have
a
method
that
allows
us
to
get
access
to
that
info.
B
For
the
current
span
to
say:
is
it
a?
Is
it
a
client,
and
I
think
that's
all.
C
B
Or
even
the
the
full
I
mean
now
now
we
can
use
readable
span
because
get
kind
is
on
there.
Now.
C
B
Yeah,
even
if
we
didn't
have
that
we
could
cast
to
the
full
or
quick,
can
we
even
is
it
packet?
Maybe
the
full
one
is
packed.
It's
packaged
private.
So.
B
No,
that's
great,
then
that
get
kind
is
accessible.
Then.
C
B
Only
that
well
yeah
there's
so
the
main
thing
is:
it
doesn't
like
it's
outside
of
the
spec
sort
of
of
sort
of
standard,
but
it
does.
There
is
a
something
that
actually
like
our
interop
would
work
better.
If
we
did
it
this
way
like
if
users
manual
instrumentation,
does
client
spans,
then
that
we
would.
C
B
Yeah,
and
and
even
even
if
we
did
interrupt
with
that,
key
users
still
have
to
know
to
set
that
context
key
and
it's
not
part
of
the
official
spec
sdk.
B
B
B
It
was
off
by
default
anyway,
but
it
would
create
spans
for
each
filter
call,
and
so
the
problem
is
in
this:
the
mvc
framework,
the
get
parent
may
not
be
the
server
and
I
guess,
there's
nothing
stopping
users
from
creating
spans
in
their
filters
on
their
own.
B
So
the
question
is
in
the
mvc
in
the
controller
method,
are
we
comfortable
assuming
that
the
parent
is
always
the
server
or
not
so
by.
B
The
oh,
the
controller
is
an
internal
span,
but
I
mean
the
when
we
update
the
route
update
the
span
name
of
the
server
with
the
route.
Do
we
just
get
parent?
I
guess
we
can
check
and
say.
If
the
parent
is
a
server
span,
then
we
update
it.
If
not,
we
kind
of
say
too
bad.
C
Do
I
mean
I
tried
very
hard,
hoping
there's
some
instrumentation
way
to
do
it
without
that?
That's
why
I
went
through
that
whole
process
on
that
pr
with
you,
but
I
think
we
found
that
it's
not
possible
generally
to
do
it
in
instrumentation
right.
We
do
need
that
key
and
if
that's
the
case,
it
is
what
it
is.
I
think.
B
B
C
B
Okay,
so
I
think
probably.
B
To
make
bogdan
happy
wish,
we
should
open
a
spec
issue
about
our
need
for
access
to
the
local
root
span.
So,
okay,.
B
Yeah,
cool
yeah
I'll
have
to
kind
of
go
through
exactly
remember
exactly
why
we
can't
work
around
well.
I
think
that
there's
certainly
that
filter
example,
but
the
jax
rs
thing
remember
what
these
jacks.
C
C
B
Cool
more
ideas
from
bogdan
a
lot
lots
of
ideas
from
baghdad.
We
told
at
the
end
I
think
we
we
agreed
that
he
can
join
army.
He
should
can
and
should
join
our
meeting
once
a
month,
one
not
too
often
or
we'll
always
get
derailed,
but
often
enough
that
he
definitely
has
good
thoughts.
B
The
tracer
name.
Oh
yeah.
He
was
saying
that
it
was
confusing
to
him
that
we
call
our
things
tracers
and
also
he
was
suggesting,
since
we'll
probably
want
to
them
to
include
metrics
of
renaming
them
to
instrumenter
or
some
something
else
from
tracer.
C
A
C
B
Yeah
yeah
so
yeah,
it
makes
sense
to
me
to
just
if
it's
just
renaming
some
things
thinking
ahead,
even
if
it
even
if
we
still
break
stuff,
if
we
break
less,
I
guess
that's.
C
B
C
B
A
This
is
so,
I
know
where
you're
going,
and
this
is
actually
a
good
question,
because
I
think
baggage
in
open
census
and
maybe
in
open
tracing,
I
don't
know
open
tracing.
You
could
put
metadata
on
where
what
the
life
cycle
of
that
particular
baggage
item
should
be.
That's
the
usage
of
the
metadata-
and
you
can
say
it's
only
scoped
to
this
particular
service
and
then
it
will
not
be
propagated
outside
the
service.
C
A
B
C
A
Yeah,
so
I
guess
that's
what
packages,
but
I
think
I
know
that
christian
has
suggested
that
context,
maybe
should
be
able
to
speak
through.
C
C
B
Yeah,
I
would
like
kind
of
like
to
put
the
route
or
the
local
root
span
name
into
the
context,
because
our
back
end,
our
our
older
solutions,
would
stamp
the
we
called
it
operation
name,
which
was
the
span
name
of
the
request.
B
It
would
get
stamped
on
all
of
the
like
the
log
data
and
the
outgoing
span
data
and
we'll
see
so
far.
We've
had
one
customer
ask
where
that
feature
went.
B
No,
it's
just.
It's
called
the
operation
name
and
so
the
like
trace
data
client
span
data
have
their
own
span
name,
but
in
addition
to
that,
we
stamped
them
with
the
operate.
The
overall
operation
name.
A
C
C
B
B
And
then
this
I
was
wanted
to
ask
you
about
this,
so
bogdan
mentioned
that
he
had
had
problems
time
out
problems
with
sona
type,
also
in
another,
some
other
project,
and
he
said
that
he
had
what
he
had
found.
Helped
was
disabling
parallel
builds
during
the
publishing,
which
made
kind
of
ring
a
bell
like
hey.
Maybe
that
could
help
even
with
bin
tray
or
anything.
C
C
C
B
B
Yeah
so
yeah,
so
that
was
today
john
yous,
you
mentioned
chat
you
and
nikita
chatted
yesterday
about
spam
processors.
A
Yeah
we
were
just
I
mean
basically,
he
and
I
were
the
only
ones
in
the
sig
meeting
yesterday.
It
was
very
small,
literally
just
the
two
of
us,
so
we
were
just
talking.
We
were
mostly
just
kind
of
brainstorming
and
rambling
about
configuration
about
sdk
configuration
and
brought
up
the
question
of
like
and
alright
you-
and
I
have
chatted
about
this
as
well.
A
But
then
you
have
it
more
of
a
more
robust
exporter
framework
that
you
can
give
any
exporter
can
can
request
a
batcher
or
have
it
be
configured
with
a
batcher
that
actually
handles
the
batching
logic
rather
than
having
the
spam
processor
handle
the
bashing
logic
in
the
batchband
processor
case
anyway,
we
were,
it
was
just
really
just
brainstorming
and
it,
I
think,
we're
all
realizing
that
the
like
one
of
the
things
that
I
know
that
people
have
been
talking
about
wanting
to
do
is
be
able
to
chain
span.
A
B
A
You
would
want
to
be
able
to
make
sure
that
runs
before
the
spam
processors
actually
is
doing
export,
because
if
you
do
it
after,
if
it
runs
after
the
one
that
does
export
it
does
no
good
at
all.
So
you
want
to
be
able
to
actually
build
like
a
processor
chain.
A
C
B
C
To
add
them
yeah,
because
then,
if
you
already
have
an
exporter,
then
you
call
that
and
then
who
knows
what's
going
on?
I
think
that's
related
to
that.
If
you
always
have
to
set
your
span
processors,
when
building
at
least
then
the
order
is
still
sort
of
obvious
and
it's
better
than
what
we
have
now
where
you
can
have
spam
process.
Whenever
you
want
them,
then
the
ordering
gets
very
confusing.
A
So
anyway,
the
the
collector
has
all
this
capability
I
think
built
into
it.
There's
just
some.
You
know,
chatter,
that
it
might
be
nice
to
figure
out
how
to
build
a
more
robust,
pipelining
solution,
but
obviously
this
is
not
something
that's
happening
right
away.
We're.
A
Yeah,
I
mean
there's
lots
of
ways
you
could
yeah,
you
can
cook
it
up.
I
don't
think
there's
any.
That
necessarily
are
obviously
the.
C
A
So
anyway,
that
was
mostly
what
we
were
talking
about
is
right.
Now
you
can
configure
your
exporter
by
just
wrapping
it
in
your
own
spam
processor
and
configuring
that,
and
it
works
fine.
So
my
goal,
at
least
with
this
set
of
prs
that
I
put
the
last
one
in
today-
was
to
make
sure
we've
got
the
basics,
and
then
we
can
start
figuring
out
how
to
get
fancy
after
we're
happy
that
the
basics
are
fine,
yep
yeah,.
C
A
A
A
So
I
wanted
to
mention
that
jason
and
I
are
going
to
be
pairing
on
building
some
instrumentation
which,
because
we
have
a
customer
request
to
build
some
instrumentation,
so
we're
going
to
prepare
and
try
to
figure
out
how
to
do
it.
It's
kind
of
a
crazy
one,
though,
in
my
I
like
I,
I
saw
the
story
in
our
jira
and
I'm
like
wait.
What
they
want
instrumentation
around
the
hibernate
validator,
which
feels
like
a
very
strange
thing
to
have
instrumentation
around.
A
Oh
yeah,
so
anyways
we're
going
to
be
working
on
that.
Well,
if
you've
done
that
before
then
you
may
have
some
understanding
of
what
like,
if
there's
gotchas
in
there,
but
my
guess
is
they
just
want
to
be
able
to
put
make
sure
they
have
timings
around
that
validate
method
on
the
val
on
the
validator
api,
because
of
course,
users
are
going
to
make
their
validators
call
out
to
external
services
or
do
all
sorts
of
other
insane
things
right.
B
A
Well,
anyway,
you
don't
have
to
you,
don't
talk
about,
but
just
that
he
and
I
are
going
to
start
working
on
this.
I'm
guessing
we've
we're
starting
to
build
a
sample
app
and
we'll
probably
start
trying
to
figure
out
how
to
do
instrumentation
soon.
So
that's
our
first
job.
B
A
B
You
it's
just
a
list
of
it's
just
in
glow
root.
It's
a
list
of
json
point
cuts
that
define
you
know
what
methods
get
instrumented,
which.
A
A
Do
you
know
so,
do
you
know,
like
the
thing
that
we
were
trying
to
figure
out,
is
trying
to
understand
why
this
is
a
thing
you
want
to
instrument?
Why
are
they
what
like,
what
makes
them
slow?
Is
there
something
we
can
do
to
help
people
understand
why
their
validators
are
slow
more
than
just
kind
of
putting
a
span
around
the
validate
method.
B
What
you,
what
and
what
I
think
people
want
to
know
is
which
validators
are
the
slow
ones
and
how
many
times
you
know
are
they
getting
run
because
they're
getting
run
on
during
one
request?
You
may
run
this
one
validator
across
just
way
too
many
hibernate
objects
also
right.
So.
B
The
count
and
the
duration
and
the
name
of
the
validator
and
really
all
you
care
about
is
the
metrics
like
it
would
be
way
overkill
to
capture
spans
span
data
for
that.
B
But
we're
keeping
we're
going
to
keep
them
in
the
in
the
java
agent.
A
Okay,
the
other
thing
we
could
do
is
we
could
add.
We
could
add
events.
We
could
add
spam
events
for
that
for
validator
activities.
It's.
A
A
Interesting,
all
right,
that's
good
to
know.
I
will
chat
with
jason
some
more
given
that
info
yeah.
We
don't
have
any.
We
have
literally
like
our
requirements,
our
instrument
spring
or
hibernate
validator,
I'm
like
all
right
well
more
info
would
be
useful,
so
we're
going
to
hopefully
reach
out
to
the
customer
or
to
the
sales
engineer
and
find
out
like
what
they're
really
interested
in,
but
your
that
that
info
is
useful.
A
B
A
Yeah,
well,
I
think
idea
maybe,
but
don't
touch
the
project.
What's
that,
I
think
idea
may
be
done.
Indexing
the
project
by
starting.
B
This
meeting-
yes,
yes,
but
keep
manning
the
the
sdk
for
us.
It's.
A
My
real
hope
is
that
I
will
I
can
groom
jason
to
kind
of
also
be
a
third
maintainer
to
help
out
take
a
little
bit
of.
Allow
me
to
do
a
little
bit
more
instrumentation
work
and
offload
a
little
bit
of
the
the
west
coast.
Sdk
work.
B
Cool,
so
that
that's
actually
good
now
that
you
can
break
ties.
A
B
Well,
that's
true,
but
that's
why
you
should
have
three
people
and
then
the
you
you
can
break
ties.
It
just
takes
two
people
to.
A
A
You
were
working
together
before
in
europe,
oh
long
before
new
relic.
No
I've
hired
him
twice
as
his
manager
at
various
companies
and
had
him
like
also
at
least
two
two
times
he
hired
me
once
anyway.
Yeah
we've
worked
together
a
long
long
time,
like
literally
I
think,
15
years
since
2005.
B
A
A
My
manager
was
like
don't
hire
jason
away,
I'm
like
I'm,
not
gonna
recruit
and
I
didn't
recruit
him.
A
C
A
B
And
you're
acquiring
like
like
open,
telemetry
contributors,
went
right
and
left.
A
C
A
B
You
guys
are
all
like
the
stuff
nikita.
You
know
just
pushed
with
all
the
matrix
smoke
tests
app
server
stuff.
Like
basically,
you
know
you
guys
are
super
dedicated
to
up
streaming.
You
know
exactly.
C
A
On
work
they
did
at
plumber,
so
they
had
this
at
plumber,
with
a
huge
matrix
of
all
their
different
app
servers
and
frameworks
that
they
had
tests
run
against.
So
they
have
a
quite
the
group.
There
has
a
lot
of
experience
in
building
and
hopefully
evo
who's
there,
who's
like
their
ceo
or
cto,
or
something
and
he's
now
a
product
manager
at
splunk
he's
the
instrumentation
product
manager.
Hopefully
he
will
start
coming
to
meetings
also
to
talk
about
so
he's
started,
showing
up
to
the
spec
meetings.
A
A
B
You
know
in
our
distro,
but
it
takes
time
and
effort
to
push
all
that
stuff
upstream
or
you
know,
and
to
do
it
upstream.
So
that's
where
you
know
that
yeah
I
recognize
that
that
is
not
not
not
for
free
in
a
big
help
for
all
of
us.
B
Android
was
there
anything
you
wanted
to
else
you
wanted
to
chat
about?
Are
we.
C
A
A
A
C
A
I
said
I
know
in
principle,
but
I
I
I
don't
have
a
cs
degree.
So
I
don't
know
how
any
of
this
stuff
actually
yeah
and
I
do,
but
I
don't
know
what's
going
on.
I
don't
think
it's
about
the
csv.
When
I
took
cs
classes,
we
were
still
learning
about
heap
and
stack,
and
these
fancy
data
structures
were
in
the
correct
yeah.
C
A
A
B
Where
it
will
so
that,
because
right,
if
you
can't
keep
adding
the
same
thing
to
it
over
and
over
right,
you
can
get
a
really
big
one,
and
I
remember
bogdan
explaining
this
at
one
point
that
at
google
they
had
some.
But
you
know
like
a
memory
explosion,
bug
where
people
were
just
putting
way
too
much
in
there,
and
so
some
of
the
optimizations
in
here
work
to
deal.
C
With
that,
okay,
then
that
makes
a
lot
more.
So
what
I
was
imagining-
and
maybe
what
john
was
imagining
was
for
the
persistent
was
you
keep
on
adding
to
the
try
and
that's
why
you
don't
have
to
copy
stuff
in
the
normal
way
and
that's
where
you
can
have
the
memory
explosion?
So
even
if
you
overwrite
a
value,
you'd
still
add
it
and
still
have
the
rest
of
your
try
or
the
rest
of
your
tree
underneath
it.
C
C
A
Yeah
jason-
and
I
were
talking
a
little
bit
about
like
he-
put
some
comments
on
the
pr
with
the
array
right
and
I
was.
I
started
wondering
like
how
many
like
how
like
you
you
with
your
benchmark
when
you
got
above
five,
it
started
the
tree
started
winning
or
try
started
winning
right
yeah.
So
the
question
is:
what
is
a
typical
number
of
contexts?
C
A
Yeah
yeah,
I
mean
it.
B
In
auto,
instrumentation,
currently
is
typically
just
three
server
internal
for
the
spring
control
like
a
controller
span
and
then
the
client
on
the
way
out.
B
Stuff
that
I'm
not
really
that
big
of
a
fan
of
that
creates
a
bunch
of
internal
spans,
but
still
it
would
just
be
one
level
one
more
level.
C
Even
though
john
it's
not
the
number
of
elements,
that's
going
up
because
we're
replacing
the
same
key
just.
C
Yeah,
so
we
like,
I
think
I
would
expect
a
context
in
most
apps
to
have
no
more
than
five
elements
inside
it
yeah
like
actually
we're
talking
about
removing
the
client
spam
piece,
so
we
might
even
go
down
to
four
or
something
but
right
now
we
have
the
spam.
The
baggage
server
span.
Client
spin,
so
maybe
is
that
all
we
do.
I
think
that
might
actually
be
all
we
have
right
now.
C
A
A
C
C
Right
and
so
the
instead
of
the
infinite
tree,
which
is
the
only
way
I
could
think
of
implementing
a
persistent
structure,
they
have
this
compaction
step
to
prevent
it
from
being
infinite.
So
that
makes
sense-
and
that's
probably,
why,
like
in
our
contrived
benchmark,
the
infinite
thing,
would
have
the
same
memory
usage
but
be
much
faster.
The
reason
it's
so
slow.
I
think
this,
because
this
compaction
actually
had
so
much
overhead
that
it's
just
not
really
worth
using
this
complicated
thing.
Anyways
is
my
sort
of
gut
instinct
right
now.
B
This
thing
does
stand
out
as
a
as
a
something
something
odd
when
I
had
asked
some
folks
internally
at
microsoft,
to
review
the
that,
when
I.
B
With
yeah
the
the
the
context
propagation
stuff,
when
I
had
extracted
it
out,
this
was
the
they
were
like
that
looks
like
a
smell
like
like
normally
like
you
know,
some
massively
complicated
data
structure.
B
C
C
Yeah
and
like
your
language
probably
can't
handle
it
optimally,
anyways,
even
if
the
numbers
in
the
paper
look
good,
that's
what
I
think
is
happening
here,
yeah
I'll
be
happy
to
like.
I
don't
like
having
code
that
I
don't
understand
very
well
here.
I
trust
it
because
it's
in
the
grpc
context,
but
it
might
not
be
the
best
thing
we
have
anyways.
B
Depending
on
whether
you
want
to
or
not
I
mean
bogdan,
certainly
understands
all
of
this,
he
was
he
patched
he
made
when
we
were
doing
the
context
work
when
we
were
you
know
extracting
it,
he
had
patched
the
grpc
context
in
their
repo
in
a
few
places
earlier
this
year,
so
yeah
he's
definitely
got
all
the
this.
I
think
this
was
his
code
in
google.
I
think
he
owned
all
the
context,
propagation
stuff,
but
yeah,
and
it
definitely
I
think
his
perspective
on
it
is,
is
based
on.
B
You
know
a
lot,
some
of
a
few
weird,
really
weird,
like
rare
cases
of
people
putting
you
know
thousands
of
things
into
context,.
C
B
C
B
Yeah
I
mean
if
you
look
at
the
the
reactor
context,
I
mean
it's
not
even
a
tree,
not
even
a
try,
it's
just
a
map
and
it
just
gets
copied
each
time.
Yeah.
B
Oh
yeah,
I
was
just
getting
this
afternoon,
we're
refreshing.
C
C
C
A
C
The
unit
tests
are
obviously
much
nicer
with
having
them
right,
and
we
have
like
this
so
spawned
mainly
from
that
issue,
that
carlos
filed
that
the
baggage
should
have
iterable.
And
then
he
added
a
comment
that
if
we
added
the
baggage,
we
should
add
it
to
everything.
And
I
agreed
with
that
comment
and
added
as
map
yeah.
A
A
Well,
I
think
the
bigger
thing
is
that
right
now,
there's
a
giant
open
issue
about
whether
we
should
have
labels
or
whether
we
should
just
use
attributes
for
matchmaking
labels
may
completely
go
away.
My
guess
is
that
it
will
go
away,
although
bogdan
is
fighting
tooth
and
nail
to
keep
it,
but.
C
C
A
A
I'm
not
opposed,
I
said
I
just
wanted
to
make
sure
we
were
thinking
through.
Do
you
think
you
want
to
get
that
stuff
in
for
12
or
does
anything
it
matters
very
much?
It's
not
a
breaking
change
right,
so
we
could
add
it
yeah.
C
C
A
A
C
A
C
A
A
A
C
A
A
A
A
C
A
All
right,
let
me
just
put
the
approval
on
this
one,
so
we
can
get
that
one
in
if
you
yeah,
if
you
want
to
run
a
release
tonight,
that's
totally
fine
with
me.
A
A
And
we
would
want
to
use
that
I
think,
as
the
basis
of
the
release
notes
as
well
so
and
hey.
If,
if
carlos
isn't
doing
the
release,
maybe
we
can
actually
remember
to
do
release
notes
this
time
because
carlos
never
does
them.
A
I
was
like
I
gotta
figure
out
how
to
get
release
notes
out
the
next
morning,
so
yeah.
C
B
A
B
B
Did
you
want
to
chat
anything
else?
Dm?
Oh,
I
was
going
to
ask
the
so
the
interfaith
the
the
span,
context
and
stuff
like
that,
for
you
were
asking
about
0
12,
because
we'll
have
to
update.
C
No,
not
personally,
no
just
wondering
okay
yeah
yeah
well,
technically,
if
you
were
to
release
12
next
week,
then
this
way
doubles
is
going
to
have
our
ga
open.
Tell
me
whatever
that
that
means,
but
we're
launching
it
this
week.
I
guess
I'm
sorry,
not
just
we're
launching
it
in
two
weeks,
so
I
was
sort
of
planning
on
launching
point
one
one.
Otherwise
I'd
punch.
I
had
launched
point
twelve
if
that
was
out
at
the
time
and
it
really
doesn't
matter
because
neither
of
them
are
one
so
yeah
yeah,
yeah.
C
B
C
B
C
C
B
C
B
C
B
B
C
B
C
C
B
For
from
this
with
the
jumping,
oh
so
then
you
could
yeah.
I
see.
B
B
Yeah,
for
some
reason
I
was
thinking
this
would
be
a
this
would
be
different
because
it's
it's,
it
is
sort
of
different,
but
it's
still
just
instrumentation.
C
B
Right
right,
because
it
is,
I
mean
the
key,
is
it
has
to
be
on
the
system
class
path,
not
just
in
your
war
file
or
something
but
yeah.
C
B
Yeah
our
smoke
test
should
work
awesome,
okay,
cool.
I
will
test
that
out
in
the
in
the
branch.
Well,.
C
B
Will
be
good
exercise
for
me
to
understand
exactly
what
you
did
for
the
flow
of
the
span
data
not
the
same
for
metrics.
C
B
C
C
No
so
span
we're
using
the
tlp
to
switch
from
shaded
span
data
to
non-shaded
stand
data
through
that
annoying
converter.
All
right,
because
we
have
all
these
assertions
already
built
for
spam
data
for
metrics.
We
could
probably
just
search
directly
on
the
otp.
That's
probably
simpler,.
B
C
C
C
B
C
Yep
but
that's
what
I
always
thought
and
then
you
showed
me
the
way.
Remember
that
conversation
because
a
couple
months
ago,
but
like
you
in
your
in
your
microsoft
branch,
you
had
one
place,
you're
doing
that
getting
the
reference
page
and
classified
and
then
doing
something
with
it.
I
was
like
I
learned
an
awesome
trick
that
they
use,
which
I
use
for
this
bridge.