►
From YouTube: 2021-09-03 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).
B
I
believe
at
least
one
of
them
was
delayed
to
for
our
resident
grpc
expert.
B
I
forget
what
the
other
one
might
have
also
been
another
grpc.
No,
it's
probably
something
else
we'll
get
we'll
find
it.
B
A
B
I
didn't
know
that
yeah,
yes,
yeah!
No,
that
she's
been
a
regular
person
that
I've
noticed
has
gotten
pinged
on
like
cncf.
I
guess
mostly
related
to
because
there's
all
there's
been
a
lot
of
zoom
issues
in
the
past
kind.
A
Of
surprised,
she's
still
doing
that
because
she,
just
literally
like
last
week
or
two
weeks
ago,
got
her
her
jd,
so
she's
a
she's,
an
official
lawyer
now.
B
Oh
cool,
maybe
I
wonder
if
she's
doing
like
open
source
law
or
something
yeah,
no.
A
B
It
is
food
getting
into
food
time,
late,
bounce
band
process,
so
we
talked
a
bunch
this
morning
about
native
images,
graw
vm
quarkus
both
for
non-agent
and
for
agent
and
so
sort
of
the
the
ben
sort
of
outlined
his
his
thoughts
on
this
being
able
to
annotate
something
or
basically,
we
need
some
way
to
separate
out
things
that
should
be
delayed
until
run
time,
some
methods,
so
the
methods
that
primarily
binding
to
sockets
so
the
grpc
exporter.
C
B
B
A
A
Like
yeah,
I
mean
my
main
like
this:
annotation
idea
of
benz
would
be
like
it
doesn't
do
anything
under
normal
circumstances,
but
something
that
the
growl
or
corpus
or
whatever
the
pre-processor
could
rock
to
say.
Yeah,
don't
execute
that
thing,
because
it's
going
to
make
network
connections
or
do
something
else
to
be
impossible
to
deal
with
during
that
free
couple.
When.
A
Well,
we
like
the
regular
code
path.
We
just
call
it
like
if
we
just
call
this
back
normally
exactly
the
way
it
is
yeah,
but
it
would
just
be
a
hook
for
drawl
or
whatever
to
say.
Oh,
don't
call
this
thing
when
I'm
trying
to
gather
all
the
byte
code
and
gather
all
the
all
of
the
code
paths
that
might
need
to
be
instrumented,
it
might
need
to
be
pre-compiled.
B
D
C
B
A
A
A
A
A
D
Like
auto,
so
we
have
auto
configure
dot,
initialize
or
something
like
that
method.
I
can
imagine
being
delayed,
because
that
has
no
state.
It's
just
a
static
method.
You
call
but
any
method
within
that
tree
like
it
depends
on
some
state,
that's
in
memory,
and
I
didn't
think
that's
what
graphium
does
like
save
state
like
a
emulator
or
something.
I
thought
it
don't.
D
C
B
Could
be
way
off,
but
what
I
thought
he
was
explaining
the
other
day.
Oh
yeah,
because
we
chatted
on
tuesday
was
that,
like
class
initialization
also
happens
during
pre-compilation,
which
would
mean
that
there
is
state
which
it,
which
means
it
is
saving
memory
off,
which
means
that
it
maybe
it
does,
have
those
pointers
and
could
call
instance.
E
C
B
I've
I
volunteered
to
you,
asked
for
volunteers
to
pair
on
trying
to
add
these
annotations
or
something
in
so
I
will
hopefully
learn
more
and
give
you
more
correct.
Less.
E
A
Ben
evans,
we
like
too
see
if
you
can
get
it
working
wow.
Well,
he
yeah.
He
just
started
at
red
hat
a
month
or
so
ago,
and
it's
kind
of
focusing
now
on
aot
with
growl
and
corpus
and
agents
like
spinning,
up
agents,
handling,
free
main
handling
pre-made
with
with
aot
wow
yeah.
B
B
Probably
not
understanding
some
of
the
complexities
because
it
seems
to
me
with
the
pre-main
like
there's
not
really
too
much
different
like
as
long
as
during
that
pre-compilation
I
mean
as
long
as
during
that
native
image
compilation
process,
you
run
pre-main.
First
right,
we
hook
up
the
class
file
transformers
and
then
you
know
you
run.
Basically
you
have
to
run
your
app
in
order
to
dump
all
the
class
files
to
get
native
image
compiled.
D
B
D
A
A
B
I
think
the
idea
is
that
frameworks
that,
like
quarkx
for
frameworks,
can
give
you
most
of
those
code
paths
but
there's
there's
gaps,
and
so
that's
why
they
have
to
run
and
exercise
the
app
to
fill
in
those
gaps.
B
G
F
F
So
anything
you
use
reflectively
in
brawl,
you
basically
tell
it
about,
and
then
it
makes
that
like
an
entry
point,
I
know
that
so
it'll
like
flush
out
that
code
path
and
have
it
be
an
entry
point
that
you
can
call
and
that's
a
pain
in
the
butt
like
every
time
I
implement
something
with
growl
it's
in
scala
and
if
you're
not
familiar
with
scala.
F
The
lazy,
vowel
concept
uses
reflection
all
over
the
place
unsafe
for
efficiency
reasons,
and
it's
a
it's
a
pain
in
the
butt
to
deal
with
any
scala
thing
and
growl,
because
you
have
to
like
go
through
and
annotate.
Just
all
of
these
internal
details
that
you
didn't
want
to
care
about.
F
But
that's
like
that's
like
built
into
growl
right
now,
where
you
basically
run
so
so
the
easiest
way
to
use
grail
is
you
just
run
your
app
and
there's
a
there's,
a
agent
that
you
run
that
dumps
all
your
annotations
from
the
live
running
java
app.
Then
you
take
the
output
of
that
and
you
use
that
as
your
config
for
gall
vm
and
then
it
goes
and
builds
your
image,
but
usually
it's
way
too
aggressive
right
and
you
can
actually
prune
the
output
of
that
a
lot
more.
F
F
F
The
30
meg
executable,
but
then
it'll
be
nice
yeah.
We
we
gave
a
talk
at
the
last
scala
days
with
open,
telemetry
native
image.
It
was
how
many
megabytes
it
was.
It
was
pretty
tiny,
it's
pretty
happy.
Oh
yeah,
yeah
yeah.
No,
we
used
the
straight
api
and
we
instrumented
ourselves,
but
yeah
it
was.
It
was
nice
and
tiny.
B
F
B
D
B
Yeah,
because
pro
guard
is
a
pain
I
I
had
in
chicago
yeah
and
here's
all
here's
all
of
my
exceptions.
D
F
Pro
guard
still
used
didn't
that
stop
getting
supported
like
eight
years
ago.
A
A
D
A
F
So
then
I
don't
I'm
not
here's
an
issue
from
2022,
no
sorry
from
when
2022
from
last
year,
not
from
next
year
of
how
it
didn't
support
gzip.
F
It
is
marked
as
closed.
I
don't
see.
E
A
D
F
Like
it
looks
like
there
was
a
fix
to
the
collector
submitted
a
while
ago
so
september
of
last
year,.
D
D
F
17
is
the
one
with
the
new
fiber
rotini
stuff
out
in
there.
What's.
D
A
F
A
A
So
ben
was
actually
saying
last
I
don't
know
sometimes
either.
I
talked
to
him
on
the
side
or
was
in
a
meeting
a
couple
weeks
ago
and
said
the
drawl
team
and
the
jb
and
the
hotspot
team
are
not
friendly.
A
D
I
mean
yeah
is
crawl,
keeping
up
with
the
javas
like
if
it
takes
them
a
year
to
get
java
17
support,
that's
always
going
to
be
their
limiting
factor.
I
guess
especially
if
springs
doing
this
I
know
before
they
were
not
on
java
11
for
a
long
long
time.
I'm
wondering
how
their
version
could
probably
swing
nowadays,.
A
D
D
Like
I
think,
groovy3
probably
does
support
it.
It's
just
migraine
even
groovy3
is
probably
annoying.
Maybe.
B
F
A
A
F
B
A
F
F
Not
in
c
plus
it
is
old,
but
again,
like
I've,
been
writing
scala
for
so
long.
We
have
we
have
we
have
we
got
everything
in
match
statements,
man,
you
can.
You
can
like
basically
write
a
turn,
complete
match
statement
that
just
never
completes.
You
know
what
I
mean.
F
That's
that's!
That's
when
you
know
you've
won
right
when
you
can't
tell
the
performance,
but
honestly,
actually
I
actually
really
really
am
looking
forward
to
pattern
matching
hitting
java
there's
so
many
places
that
would
like
clean
up
the
code.
We're
writing
anyway.
Switch
swish
expressions
are
also
good.
I
I
realize
now
what
that
means
in
java.
A
D
A
A
F
Are
insane
yeah,
oh
speaking
of
which
record
is
reserved
the
word?
Do
you
need
us
just
to
start
changing
the
java
api
to
not
use?
The
word
record
like
it
looks
like
they're,
not
using
a
in
context,
keyword
reservation,
which
just
annoys
the
hell
out
of
me
as
a
language
design
person
I
think,
but
you
can
still
have.
We
don't
have
any
any
variables
we
can
record.
Though
dude
we
have
methods
named
record.
I
think
that's
fine.
I
don't
think
that's
a
problem
as
long
as
that's
fine,
that's
cool.
It's
just.
A
F
E
F
F
So
the
my
cat
is
climbing
on
my
computer
and
I
cannot
see
anything
right
now.
Do
you
see
this
look
at
her?
She
wants
my
milkshake.
Oh
I
mean.
F
Okay,
so
bag
engine
tracing
basically
views
right.
Now
we
specified
a
thing
on
views
that
lets
you
pull
keys
out
of
baggage
and
throw
them
onto
attributes
on
metrics
right.
F
We
actually
have
a
completeness
between
the
two
signals
of
you
know:
metrics
have
attributes
sorry,
metrics
have
baggage
into
attributes
as
a
configurable
thing
and
then
also
spans
and
spans
can
do
the
same
thing.
I
don't
know
if
that's
the
right
hook,
because
I'm
the
least
comfortable
with
my
knowledge
of
what
to
throw
on
span
attributes
in
general,
because
I
am
used
to
not
throwing
anything
on
spam
attributes
in
my
actual
personal
usage
of
tracing,
which
is
kind
of
lame,
but
that's
what
I
did
anyway.
A
So
I
think
this
is
something
that
has
been
missing
from
this
deck
for
a
long
time.
I
don't
think
I'd
want
to
put
it
into
the
main
codebase
unless
it
was
in
the
incubator.
An
incubator
is
fine
facing
the
incubator
is
a
good
place
for
it,
because
I
think
it
needs
to
go
on
respect.
A
Well,
I
think
doing
it
on
the
spam
processor,
on
onstart,
where
you
have
the
context
and
you
can
grab
the
baggage
is
seems
to
me
enough,
like
you
know,
you
could
during
the
lifetime
of
this
fan,
do
add
more
stuff
to
baggage,
but
I
don't
think
we
should
miss.
I
don't
think
that's
necessarily
a
use
case.
We
should
have
to
deal
with
right.
F
Now,
right,
I
sorry
I
meant
in
the
configuration,
so
if
we
configure
what
baggage
attributes
go
into
what
spans
there's
like.
Do
you
allow
people
to
filter
on
keys
or
or
baggage
keys
to
include,
and
then
you
also
let
them
filter
on
like
existing
spam
attributes
as
well,
for
when
to
pull
in
baggage
yeah.
Those
are
like
the
two
things
that
I
ask
in
terms
of
simple
config
right,
understand
names:
why
not
spam
names
yeah
without
spam?
Is
that
a
name
an
attribute.
A
Or
not,
it's
not,
I
say
name
is
not
an
attribute.
Okay
cause
mankind
staying
kind.
F
D
F
F
F
You
don't
want
to
attach
baggage.
I
I
don't
know
there's
to
me
there's
this
question
of.
Do
I
want
to
append
baggage
on
every
single
span,
all
the
way
down
the
trace.
You
know
every
as
that
context,
propagates
or
do
I
want
to
put
it
on
just
the
outer
one,
so
so
that
use
case
of
let's
just
put
it
on
the
outer
one
is
an
assumption
I
have.
I
don't.
I
really
don't
know
whether
you
want
it
on
all
of
them
or
not.
F
You
know,
because
I,
when
I
most
the
usages
of
baggages,
the
baggage
that
I
think
of
are
around
metrics
and
are
around
a
b
testing
and
around
experimentation
when
it
comes
to
spans.
I
I
really
actually
haven't
tried
to
get
baggage
based
experimental
span
understanding,
yet
I
think
it
would
actually
be
super
useful.
It's
just.
I
was
unable
to
do
that
internally
with
the
like
crazy
things
that
we
have
that
are
not
open
telemetry,
so
I
never
tried
it.
So
I
don't
really
know
what
I
would
want.
A
F
What
I
did
previously
was,
I
tagged
the
traces
we
don't
have
traces,
no
such
thing
I
know
I
know
so
that's
what
I'm
saying
like
things
that
I
did
internally
are
different
than
guys
do
externally.
Yeah
yeah.
A
That's
why
you
went
in
the
local
route
right.
One
issue
with
the
appending
based
on
attribute
is
that
the
spam
processor
only
gets
the
initial
attributes
like
what
are
done
via
the
builder,
not
stuff,
that's
added
later,
which
could
lead
to
a
confusing
user
experience
or
they're
like
well.
I
added
this
attribute
to
the
span.
It's
not
getting
the
baggage
same
with.
F
So,
in
terms
of
next
steps,
what
if
we
so
so,
the
the
open
question
in
my
mind,
is:
should
I
make
an
sdk
extension
for
a
baggage
span
attribute
thing
and
then
should
it
share
code
with
the
metrics
sdk
baggage
pool
stuff
or
which
honestly
turning
baggage
into
attributes,
is
a
single
line
of
java
code?
So
I
don't
really
think
it
should
share
anything
besides,
possibly
names.
F
F
I
don't
know
how
frisky
you're
feeling,
but
I
would
love
if,
if
we
could
have
that
as
an
sdk
extension
in
auto
configure
that
people
could
try
to
make
use
of
at
some
point
just
just
to
see
if
we
get
adoption,
if,
if
people
start
to
use
it
and
then
leverage
that
to
push
the
spec
quicker
of
hey,
this
is
actually
a
feature.
People
need
and
use.
A
F
A
Oh
yeah,
no,
I
know
I'm
aware
I
would.
What
I
would
say
is:
let's
keep
it
out
of
auto
configure
right
now,
because
the
configuration
is
going
to
be
complicated.
We
don't
have
a
solution
for
complex
configuration,
put
it
into
the
tracing
incubator
extension
and
use
that
as
the
disc
as
the
discussion
point.
Oh
do
we
have.
F
A
so
the
you
know
how
you
have
the
meter
provider
configure
sbi.
Yes,
what?
If,
if
I
make
one
of
those
in
the
sdk
spi
for
configuring,
this
configuring,
your
span,
sdk
you're
tracing
this
okay.
Exactly
so,
I
would
like
pull
the
configuration
properties
off
of
that
and
and
fill
it
out
in
in
an
in
so
it'd
be
an
sdk
extension
for
the
raw
processor.
F
Then
there'd
be
an
auto
configure,
sdk
extension
that
would
kind
of
append
itself,
possibly
or
even
as
a
processor.
If.
F
D
So
it
also
seems
like
a
nice
place
to
map
package
just
because
it's
all
the
attributes
and
then
you
don't
have
to
care
about
the
concept
of
package
anymore.
You
just
have
attributes
you
could
one
advantage.
Is
that
then,
and
maybe
it's
not
an
advantage-
maybe
that's
why
we
don't
do
it,
but
you
could
have
the
same
configuration
of
what
baggage
gets
to
attributes
for
both
spans
and
metrics.
It
wouldn't
be
separate
configurations
and
then
what
attributes
get
into
your
metrics?
That
would
be
a
separate
configuration
if
you
need
it,
but
at
least
with
baggage.
D
F
So
I
think
we
have
to
one
of
the
problems
with
open
telemetry
is
we
have
to
be
incredibly
flexible
and
also
good
for
the
user,
yeah
and
leaning
towards
incredibly
flexible,
because
all
the
possible
use
cases
so
from
an
instrumentation
standpoint
or
the
agent
like
that
sounds
like
the
right
level
of
config
for
the
agent
to
just
out
of
the
box,
make
it
simple
from
an
sdk
standpoint
like
it
should
be
100
you
can
configure
what
the
hell
you
want
if
you
get
dirty
right,
yeah,
so.
D
D
F
D
I
for
that,
for
this
thing
I'm
talking
about,
I
think
it'll
just
be
sort
of
a
separate
implementation
within
the
instrumentation
as
well,
though,
because
that's
like
you
might
not
even
have
the
st
like
the
instrumentation
is
a
different
layer
right.
So
in
this
case
I
don't
think
we'd
be
able
to
use
the
sdk
feature,
which
is
okay,
yeah.
D
F
F
The
next
step
after
that
is
to
extend
that
into
instrumentation.
So
you
know
the
enable
disable
flags
the
instrumentation
has.
If
you
weren't
always
dealing
with
properties,
how
much
more
flexible
would
you
want
to
be
right
and
what
kind
of
configuration
would
you
want
there?
That's
that's
like
anyway.
I
think
we
have
to
walk
before
we
run
so
step.
One
is
just
get
the
sdk
configuration
in
a
structured
file,
format
that
interacts
well
with
programmatic,
config
right
and
then
step.
F
Two
would
be
all
the
stuff
that
you're
learning
with
instrumentation
things
that
people
want
to
turn
on
and
turn
off.
My
cat
is
going
nuts
over
there.
Sorry,
the
there
was
an
open
question
on
you
know
whether
or
not
the
java
instrumentation
should
have
a
different
span
for
hdb
retries
at
one
point
right-
and
I
want
us
to
have
that
answer-
be,
of
course
it
should
it's
just
off
by
default
and
it's
really
easy
to
turn
it
on
and
off,
and
it's
like
not
a
lot
of
overhead
or
almost
zero
overhead.
F
F
If
you
agree
with
that
or
not,
but
it
sounds
like
next
steps
are.
I
can
make
an
auto
configure
extension
and
an
sdk
extension
or
putting
baggage
on
spam
processor
use
that
to
have
the
discussion
with
the
spec
sig.
None
of
that
is
higher
priority
than
getting
this
metric
exporter
stuff
actually
specified,
because
that
is
a
little
terrifying
which
we
can
talk
about
later
if
you
want,
but
that
sounds
like
next
steps,
any
any
thoughts
or
anything
any
concerns.
A
My
my
go-to
workflow
at
the
moment
for
these
kind
of
experimental
things
is
if
we
think
this
is
something
we're
going
to
end
up
in
the
spec
and
should
be
in
the
spec,
let's
put
it
into
the
sdk
as
an
experimental
extension.
A
A
We're
all
scared
that
he's
he's
taking
a
vacation
to
use
up
all
his
days
before
quitting
going
to
become
a
goat,
herder
or
something.
A
F
I
noticed
I
noticed,
there's
a
concern
with
branch
protection
rules.
What
happened
there?
If
I
can
ask.
A
Oh
so
because
we
we
keep
flip
flopping
back
and
forth
between
running
we're
going
to
18
and
go
to
20..
We
have
a
different
build
checks
for
mac
os
versus
ubuntu,
and
we
recently
put
into
pr
to
change
it
over
back
to
ubuntu
20
2004,
to
run
our
ci
on
and
when
we
went
to
github
for
the
branch
protection
rules
and
removed
the
18,
we
couldn't
add
the
20
back
like
there
was
a
bug
in
the
ui
that
wouldn't
let
us
actually
add
brand.
A
A
C
B
And
that
that
that's
one
thing
that
this
accept
pr
has
worked
out
well
for
us,
we
changed
yeah,
we
change
the
jobs
a
lot,
but
we
have
this
final
job
that
everything
funnels
into.
So
we
just
have
the
one
check
on
this.
B
D
B
Yeah
so
yeah
this
is
it.
You
just
define
what
it
needs
and
then
it
flows
into
here.
A
D
B
D
B
D
H
F
F
Frustrating
to
people,
but
I
also
we
don't
want
to
remove
that.
D
B
Same
I
remember
they
were.
They
were
supposed
to
vote
in
the
gc
meeting
today,
whether
we
could
do
this
temporarily
until
the
cncf
solved
it.
Do
they
pick
meeting
notes.
C
A
A
First
item
in
the
gc,
for
today
was
what
is
avuncular
for
ants
nice.
C
D
B
I
had
to
do
it
exact
name
that
got
tried.
B
F
C
B
I
just
wanted
to
kind
of
update,
based
on
our
conversation
from
last
week
on
iraq,
instrumenter
api
conversion
called
out
jack's
good
work.
There
we're
up
to
57
out
of
99..
B
B
B
Interest
because
it's
by
reference,
equality.
B
Right
right,
there
was
nobody.
I
had
left
that
to
there
was
nobody
from
logging.
Folks,
there
discussed
this.
Everybody
is
in
violent
unanimity
that
the
url
is
horrible
and
that
what
we
are
doing
is
horrible.
B
For
both
there
sorry,
what
was
the
oh
sorry
url.
This
is
that
java,
the.
B
Yeah
yeah
everybody
had
their
own
favorite
horror
story
about
javanese.
B
That's
what
ben
was
asking
ben
thought.
B
I
think
that
that's
not
happening
because
that
didn't
show
up
in
my
profiling,
but
certainly
just
the
the
validation
and
constructing
the
object
and
calling
to
string
on
it
showed
up
in
my
profiling
big
time.
F
D
E
F
F
A
B
D
To
use
it
yeah
I
gotta
run:
let's
see,
did
you
take
a
scan
of
the.
A
It's
my
oh,
the
migraine,
marshall,
you
tilted
to
an
abstract
class,
yeah
they're,
still
singing
my
inbox.
I
totally
forgot
to
look
at
it
I'll
I'll.
Look
at
it
first
thing
anymore.
I
think
I
took
a
glance
through
it,
but
I
haven't
had
a
chance
to
dig
in
yet
I
spent
all
my
most
of
my
day
writing
or
my
last
two
days
writing
a
workshop
for
our
field
team
for
our
my
android
instrumentation
from
yeah
from
I
need
to
drink
more.
While
I'm
writing
workshops.
F
You
know
I
I
cuz
cause,
I'm
I
have
secret
powers,
I
don't
know,
no
someone
wrote
me
into
it.
They
needed
someone
who
knew
open
telemetry
to
to
chat
about
things
so.
F
F
It's
not
really
top
secret.
It's
basically
just
you
know:
otlp
protocoly
things
that
we're
doing
that
people
will
be
happy
to
hear
about
so
yeah
and
then
we
just
talk
about
how
to
adopt
open
telemetry,
as
a
group
of
like
here
are
lessons
we
learned
at
google,
here's
lester's
mother,
splunk,
everybody
loves
each
other
and
we
just
send
data
all
over
the
place
to
each
other.
It's
a
big
happy
family
come
join
us
yeah,
okay,
cool
enough.
I
was
I.
A
A
F
Oh,
I
looked
up
when
java.net
does
name
resolution
and
it's
when
you
call
the
equals
method.
D
F
Hash
map
yeah
url,
will
do
name
resolution
if
you
call
equals.
So,
if
you
put
it
in
like
a
hash
map
or
a
set,
it
will
do
name
resolution.
F
F
D
A
B
Yeah
and
then
you
know
all
the
things
that
happened
in
the
last
week,
this
one
I
thought
was
cool.
I
hadn't
seen
I
hadn't
seen
this
before
another
another
agent
trick.
I
haven't
seen
on
iraq.
No.
D
B
D
C
D
Yeah
did
it
make
it
faster.
B
C
B
C
B
B
D
B
D
F
D
D
F
It's
it's
an
abstraction
on
at
its
core:
it's
just
an
abstraction
on
the
protocol,
buffer
output
stream,
so
that
you
write
tags
and
then
it
actually
supports
json
and
supports
other
types
of
formats
as
well,
but
at
its
core
it
has
a
coded
output
stream
that
looks
like
you're
writing,
protos,
but
you
it
can
do
json
and
things.
F
F
D
F
D
F
And
it's
not
much
different.
I
actually
did
the
same
thing
in
a
library
called
sauerkraut.
It's
about
pickling.
You
know
sauerkraut
pickling
anyway,.
F
Oh,
oh,
you
don't
you're
not
familiar
with
python
pickling,
it's
just
it's
just
how
to
get
things
in
and
out
of
byte
representation
quickly.
So
you
take
a
structure,
turn
into
bites
or
take
a
structure
turn
into
json
or
whatever.
F
F
Yeah
you
pickle
and
pickle
yeah
and
my
library
is
called
sauerkraut
because
that's
the
best
kind
of
pickling
you
can
do.
You
know,
anyway,
point
being
like
that.
It's
a
simple
abstraction.
It
holds
up
relatively
well
to
efficiency,
especially
if
you're
hand
coding.
F
The
actual
calls
to
like
write
this
method
write
this
method,
write
this
method
yeah.
So
I'm
a
fan
of
that.
I
I'll
have
to
look
at
the
pr
to
see
see
what
you
have
here,
but
you
coded
output
stream
is
not
quite
right.
You
know
what
I
mean.
You
have
to
deviate
just
a
little
bit
to
make
that.
F
D
E
D
F
D
H
F
Did
you
see
jack
berg's
request
around
generating
summaries
in
the
metrics
api.
F
I
gotta
go.
I
gotta
go
in
like
five
minutes,
so
I
remember
when
I
had
that
aggregation
extension
method.
Yeah,
that's
where
I
put
summaries
in
the
specification-
does
not
specify
summaries.
We
cut
it
from
the
specification.
Only
histograms
are
supported
right
now,
so
you
get
histogram
sums
and
gauges
and
nothing
else.
You
don't
get
that
count
aggregator
and
you
don't
get
summary
yeah
and
it's
possible
to
pull
summary
from
histogram
for
a
lot
of
cases.
F
F
Maybe
we
need
to
put
that,
like,
I
can
add
it
back
in
as
an
extension,
the
aggregator
still
exists
all
the
codes
there
is
tested.
It's
just
there's
no
way
for
a
user
to
write
it
without
touching
an
internal
package.
H
F
Otlp
does,
but
if
you
look,
I
can
send
you
the
link,
but
the
proto.
We
marked
we
marked
summaries
as
basically
a
deprecated
technique
for
doing
that
level
of
aggregation
and
it's
there
for
prometheus
compatibility.
But
we
weren't
planning
to
include
it
in
the
metrics
offerings
from
otilp.
We
want
to
encourage
people
to
use
histograms.
H
D
F
Right,
I
guess
what
I'm
asking
it
if
you
haven't
heard
anything
so.
F
F
Okay
cool:
that's
helpful;
I'm
not
gonna
rip
the
code
out
for
sure,
but
we'll
we'll
fix
it
over
time.
Okay,.
F
D
D
H
F
F
Just
good,
so
the
only
the
only
possible
thing
to
expose
is
summary,
so
we'll
keep
track
of
that.
The
other
quick
question
I
had
and
then
I
gotta
go
the
kind
of
related
the
jmx
gatherer
thing
yeah
we
so
we
started.
We
started
using
that
and
I
noticed
that
the
integration
test
on
the
collector
side
downloads,
the
1.0
alpha
version-
do
we
want
to
try
to
keep
those
too
up
to
date
like
when
we
bump
our
version
and
cut
a
release,
go
update
the
collector
or
do
we
kind
of
not?
D
B
Yeah
this
was
something
that
splunk
built
and
I
guess
I
think
they
have
users,
but
so
you're
you're
using
it
that's
good
to
know.
We
were
kind.
F
Yeah
we
started,
we
started
picking
it
up
for
some,
it's
good
for
legacy,
monitoring
for
people
who
don't
want
to
use
the
agent,
but
for
some
reason,
we'll
expose
jmx
which
to
me
like
in
terms
of
terrifying
propositions
in
java.
You
know
by
code,
rewriting
is
terrifying,
but
so
is
jmx
honestly
like
they're,
both
pretty
scary.
So
I
don't
know
yeah.
So
we
we
took
a
look
at
it.
F
We
got
it
up
and
running
just
did
some
did
some
tests
and
things
kind
of
like
we
did
the
same
thing
with
the
agent
to
just
look
at
the
various
offerings
and
see
see
where
things
stand
and
we're
trying
to
stand
up,
the
jmx
gatherer
and
we
ran
into
the
1.4
jmx
gather.
It
doesn't
work
with
the
collector.
For
some
reason
we
didn't
look
into
the
details
because
it
could
be
something
on
our
end
as
well
with
how
we
configured
it.
F
F
F
F
With
the
collector
right
yeah,
they
do
test
their
integration
with
java,
so
they
should
choose
when
to
bump
the
version.
I
think
anyway,
if
you're
not
familiar
with
it
all,
that's
that's
a.