►
From YouTube: 2021-10-14 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
Yeah,
I
I
saw
you
already
approved
and
reviewed
several
prs.
Well,
I
I
do
have
something
I
want
to
discuss.
Let
me
share
my
desktop.
A
So
I
believe,
like
most
of
the
feature
freeze,
requires
things
like
the
min
max
and
trying
to
clarify
the
the,
not
a
number
and
infinite
behavior
they
already
have
pr
and-
and
you
probably
reveal
all
of
them-
there
are
two
open
issues
I
haven't
figured
out.
Should
we
include
them
or
we
just
kick
them
out
from
feature-
freeze
well
I'll,
always
plan
that
quickly.
A
So
this
is
the
number
one
thing
we
need
to
have:
a
ability
to
access
those
data.
The
second
thing
is,
they
expect
there's
some
standard
name
so
take
this,
for
example,
for
tracing
basically
we're
saying
the
the
library
name
and
the
version
should
be
tags,
but
for
metrics
I
think
we
cannot
afford
to
idols
by
default
because
of
the
memory
cost.
So
what
I
I
think
we
can
do
is
we
can.
We
can
tell
when
the
isdk
called
the
exporter
or
the
reader
with
a
batch
of
matrix.
B
Yes,
I
would
agree.
I
have
recently
gone
through
the
go
sdk
and
made
changes
in
this
area.
What
I
did
was
give
exporters
a
two
level
iterator,
so
you
can
iterate
or
iterate
over
each
library
and
then
for
each
library
you
iterate
over
each
instrument.
So
that's
how
I
handled
it
and
I
suppose
that
qualifies
for
the
language
you
use.
I
agree.
A
Yeah
exactly
so,
my
question
is,
of
course
you
can
see
like
we
can
do
it
in
a
spec
now,
but
do
we
have
to
because
even
if
I
look
at
the
tracing
spike,
I
look
at
the
history.
This
feature
was
added
after
the
initial
stable
release.
It
was
added
five
months
after
we
shipped
the
stable
version
of
the
spec,
and
I
can
imagine
for
metrics
if
we
need
that
it's
an
additive
change
not
going
to
break
anything.
B
Yeah,
I
would,
I
would
actually
maybe
go
about
it
by
saying
this.
You
may
use
these
and
the
standard
name
is
the
same
semantic
convention
used
in
tracing,
but
the
expectation
is
that
they're
off
off
by
default,
I
mean,
as
for
the
reason
you
gave
yeah
and
then
they'd
use
hotel
library
name.
If
they
really
care
lightstep
uses
instrumentation.name,
we
would
change
it.
A
Yeah,
sorry,
it
seems
I
interrupted
you
several
times.
I
have
some
lagging
on
my
side.
So
when
you're
talking,
I
couldn't
hear
that
I
thought
you
stopped
then
I
started,
then
I
thought
oh
jesus,
it's
okay,
I'm
on
the
web
app
because
something's
broken
anyway.
This
sounds
good
yeah
and
I
figure
whatever
I
do
here.
A
I
probably
can
improve
the
wording
in
the
spike
and
mention
this
a
little
bit,
but
we
won't
be
able
to
close
this
issue
because
it
has
a
semantic
convention,
part
so
I'll
ping
arming
offline
and
see,
if
he's
willing
to
split
the
issue
to
an
sdk
and
a
semantic
convention.
If
that's
the
case
might
be
able
to
close
the
sdk
issue,
if
he
would
want
to
keep
the
semantic
convention,
I'm
going
to
kick
this
out,
but
still
send
a
pr
try
to
improve
the
sdk
spec
wording.
B
Yeah
I
mean
I,
if
you,
you,
could
also
just
ask
him
if
he
wants
to
write
this
because
it
it's
basically
take
what
trace
does
and
do
it
again?
You
know
yeah
exporters
that
aren't
otlp
have
the
option
and
exporters
may
use
this.
You
know
these
names
are
submitted
conventions
he
might
be
able
to
just
solve
his
own
problem.
A
Yeah
and
the
the
next
one
I
I
think
it's
relatively
easier,
so
it's
basically
trying
to
get
some
advantage
from
micrometer,
so
we
can
simplify
the
api
call
it's
basically.
If
you
have
something
that
that
you
can
simplify
like
simply
return
a
value,
and
you
know
the
dimensions,
then
you
don't
need
the
observer
object.
If
it's
a
callback,
then
you
keep
reporting
thing
so
so
this
is
from
jonathan
and
the
idea
was
coming
from
micrometer.
A
I
believe
we
can
do
it,
I'm
not
seeing
a
lot
of
support
or
us
that
we
should
have
that.
I
believe
if
we
need
some
simply,
we
should
first
focus
on
the
stable
release
and
get
feedback.
If
we
do
see
a
good
usage
of
this,
we
can
add
that
it's
not
going
to
be
any
breaking
change,
but
so
far,
based
on
the
feedback,
I
see
here,
I'm
not
seeing
a
lot
of
support
for
this
case.
B
Yeah
I
hadn't
read
it
before
so
I'm
kind
of
skimming
it
now,
isn't
this
sort
of
customization
something
we
allow
in
a
broad
kind
of
general
language.
You
know,
I
don't
know
why
this
like,
if
java,
if
the
java
community
could
be
convinced
to
let
this
interface
exist,
I
don't
see
any
problem
on
a
data
model
level,
so
I'm
not
sure
why
we
would
be
so
strict
as
to
say
you
can't
do
that.
A
This
this
one
is
on
the
api,
so
basically
jonathan
would
want
to
have
the
the
simpler
version
of
the
api
on
top
of
the
existing
api.
It's
it's
more
like
what
tit
mentioned
as
a
convenience
layer
api.
A
So,
for
example,
if
you
know
this
is
my
sensor-
and
this
is
in
the
living
room-
and
I
already
have
the
sensor
callback,
then
why
would
I
write
another
callback
and
then
use
the
observer
function
and
then
call
the
sensor
and
then
report
the
thing
so
write
five
lines
of
code?
Then
I
can
use
one
single
line.
B
Yeah
and
then
I
mean
I
know
that
there
are
connected
conversations
in
in
go,
there's
two
pieces
of
the
asynchronous
instrument
framework
that
are
questionable
from
the
spec.
One
is
bound
instruments
and
no
java
has
those
and
the
other
is
batch
observers
and
both
of
those
sort
of
kind
of
look
a
little
bit
like
what
jonathan
is
asking
for.
B
Yes,
the
bound
api
is
is
one
of
them.
The
other
is
a
any
kind
of
batch
observer
where
I
use
it
for
mems
stats
and
go
like
you
call
mem
stats
it's
expensive
and
then
you
get
14
stats
out
of
it.
So
I'm
going
to
give
you
14
measurements
in
one
callback
and
that's
neither
spect
and
it
even
has
a
race
condition
right
now
we
gotta
fix,
so
maybe
it's
maybe
needs
to
be
removed
and
maybe
there's
a
better
way
to
do
that.
B
Okay,
the
other
feature
that
I'm
aware
of
not
in
an
hotel
prototype
but
there's
a
segment.
I
o
library
that
uses
go
structs
and
structs
have
tags
and
metadata
on
them.
So
you
can
pass
one
struct
with
all
your
attributes
in
it
and
it
has
three
or
four
different
attributes
specified.
B
That
would
be
I'm,
in
my
opinion,
a
great
way
to
do
it.
It's
just
a
different
way
and
I'm
not
sure
which
the
hotel
api
will
support
and
whether
it
will
support
optional
modules
that
let
you
do
convenience,
but
I
wouldn't
say
no
to
anything
and
if
we
can
write
the
spec
to
say
each
language
should
use
the
idioms
and
performance
techniques
it
has
at
its
hands
to
make
this
better
for
users,
then
I
think
jonathan's
issue
can
be
solved.
A
I
see
thanks
I'll
I'll
I'll
put
the
meeting
notes
later.
Based
on
our
discussion.
I
want
to
see
anything
else,
because
I
I
I
think
once
we
have
those
open,
trs
merged
here,
we
should
be
ready
for
feature
freeze,
and
I
know
a
lot
of
folks
are
at
cubecon.
I
try
to
ping
some
folks
and
I
didn't
get
to
reply,
so
I
fully
expect
that
they
won't
reply
in
probably
the
next
couple
days,
so
maybe
we
we
can
target
like
our
further
delay.
A
The
feature
freeze
date
because,
like
we
originally
were
saying,
feature
freeze
by
end
of
september
now
we
make
that
made
october
and
it
is
already
made
tomorrow
is
mid-october,
so
one
week
I'll
I'll
bring
this
topic
in
the
spec
meeting.
Hopefully
we
can.
We
can
reach
feature
freeze
this
month.
I
I
think
we
shouldn't
delay.
B
A
A
Okay
and
with
the
the
min
max
and
the
the
infinite,
not
the
number,
some
of
the
clarification
here
merged-
I
I
I
think
I
I'm
not
seeing
any
other
stuff.
B
A
It
seems
we
haven't,
got
a
good
number
of
approvals,
so
I'll
ping
people
offline,
I'm
already
doing
that.
It's
just
kubecon
like
like
delayed
the
progress
yeah
yeah
and
on
the
prototype.
Have
you
seen
any?
Have
you
seen
any
like
challenges
from
I
from
allen
is
also
here
so
from
down
at
I.
I
think
there
are
a
couple
of
things
that
we
should
address,
but
I'm
not
seeing
any
blocker
that
would
require
us
to
come
back
to
the
api.
Sdk
spec
make
the
change.
A
C
We
have
some
struggles
that
are
specific
to
python,
but
I
don't
believe
they
are
anything
to
do
with
the
metric
specs.
So
I
it's
not
blocking.
A
The
future
freeze,
I
I
I
see-
and
I
notice
you
you
haven't
been
involved
in
the
matrix
spec
like
pr
reviews,
so
probably
basically
with
python
implementation.
I
wonder
if
you
notice
any
anything
you
would
want
me
to
address
before
we
announce
the
metrics
back
as
stable,
like
historically
python
prototype.
I
want
to
get
some
feedback.
If
you
don't
have
the
answer
right
now,
it's
just
like
find
the
time
to
catch
me.
C
Yeah,
I
don't
have
the
answer
to
that
right
now,
okay,
I
haven't
been
involved
in
the
metric
specs
recently.
Only
only
now
I
have
had
the
resources
to
start
looking
yeah.
I
I.
A
Understand,
okay,
then
I
I
guess,
that's
all
so
I'll
clean,
arming,
basically
and
I'll
I'll,
put
this
callback
thing
out
of
the
scope
and
and
thinking
about
once
we
get
the
metrics
part
done.
I
probably
want
to
kick
off
github
issues.
It's
trying
to
collect
what
max
I
know.
We
talked
about
the
bond
instruments
and
also
josh.
I
know
the
exponential
histogram
seems
to
be
there
and
I
want.
I
want
the
sdk
part
people
can
configure.
A
That
means
they
need
a
better
way
to
specify
the
buckets,
and
I
also
want
to
see
because
in
the
data
model
we're
saying
these
are
two
different
things,
at
least
for
otlp,
but
I
think
for
premises.
Loading
it'll
just
be
the
normal,
explicit
buckets
those
things.
I
guess
we
were
not
trying
to
cover
in
the
initial
sdk
spec,
but
after
the
initial
release
I
I
want
I
want.
I
want
to
know
if
you
would
have
energy
and
interest
to
work
on
the
matrix
spike
version
like
1.1
or
something.
A
B
Yes
and
I
have
been
trying
to
figure
out
what's
the
right
path
to
like
releasing
exponential
histograms,
I
have
a
prototype.
Neuralic
has
a
prototype,
they're
they're
nice.
B
I
really
like
this
code,
but
we
have
to
get
it
into
the
collector
first
and
then
the
questions
that
you
just
asked
for
the
sdk
are
already
being
asked
in
the
collector,
because
you're
going
to
receive
otlp,
it
may
or
may
not
have
a
new
exponential
histogram,
because
version
0.11
has
it
and
the
collector
will
soon
have
it,
but
before
we
can
give
it
to
the
collector,
we
have
to
answer
exactly
this
question.
The
prometheus
exporter
is
about
to
is,
is
holding
in
its
hand
an
exponential
histogram.
What
should
we
do
and
there
are?
B
There
are
options,
but
there
they
need
configuration
and,
and
one
of
the
options
is
return
errors.
One
of
them
is
convert
to
a
fixed
histogram
and
then
there
are
probably
more
options,
but
I
don't
think
any
of
them
are
good
for
prometheus,
so
I'd
stop
there,
and
and
and
when
we
do
answer
those
questions,
you're,
probably
right
that
they
will
also
come
up
in
the
sdk.
A
B
Do
I
actually
created
an
issue?
I
should
show
it
to
you
it's
in
the
collector
contrib
or
I
actually.
Let
me
let
me
paste
them
into
the
notes
or
sorry,
I'm
going
to
paste
it
into
the
zoom
chat
for
now,
because
I
don't
have
the
notes
document
opened.
A
B
This
is
issue
collector
issue
5742.
Oh
sorry,
that's
the
contrib
issue,
which
is
a
response
to
a
pr
that
josh
filed
already
here.
It
is.
B
This
is
collector
issue.
4197,
alex
bowden
from
lightstep
is
going
to
pick
up
the
first
part
of
this
issue,
just
to
add
the
support
for
the
exponential
histogram
to
the
p
data
of
the
collector,
but
we're
very
soon
gonna
have
to
answer
the
question
of
what
do
you
do
when
you
don't
know
about
that
type
yeah
and
then
I'm
planning
to
get
the
exponential
histogram
code
into
the
collector
first,
the
the
shortcoming
for
me
is
and
go.
We
don't
have
a
view
mechanism.
B
Yet
one
of
the
there
are
two
major
pieces
missing
in
the
go
prototype
it's
views
and
exemplars
and
until
then
it's
easier
for
me
to
work
on
the
collector
than
to
work
on
this
thing
that
I
can't
quite
contribute
configure.
Yet,
although
I
will
say,
lightstep
has
hired
one
engineer
from
the
go
community
that
we're
gonna
put
to
work
on
some
of
this
stuff.
So
we
should
have
more
progress
on
that
prototype.
A
Then
we
can
plan
from
there
and
see
what's
the
next,
so
after
the
matrix
spec,
when
stable,
probably
I'll
work
with
the
other
folks
to
do
a
blog
post
on
medium,
and
we
should
have
the
roadmap
for
what's
next
and
and
also
I
I
I'm
I'm
guessing-
there
are
some
something
related
to
matrix,
but
we
we
didn't
even
start
with
again,
for
example,
I
remember
on
the
spike
issues.
There
is
one
ask
about
the
raw
measurements,
because
people
want
they
have
some
system
they
want.
A
They
want
us
to
have
some
exporter
or
processor.
They
can
because
the
long
measurements.
I
guess
this
is
probably
for
some
alert,
for
example,
because
if
I
have
a
hdb
duration
or
more
than
100
seconds,
then
it's
crazy.
I
want
to
take
a
crush
down,
so
that's
that's
something
we
we
didn't
even
try
to
cover
in
in
this
entire
year
right,
there's
others
and
people
are
saying.
A
I
already
got
some
metrics
from
somewhere
like
another
library
and-
and
I
have
the
timestamp
already
I
I
don't-
need
the
open,
telemetry
sdk,
to
give
me
a
time
stamp.
I
know
what
I'm
doing
just
go
and
and
read
it
or
export
that
currently
I
think
we
don't
have
that
support,
although
people
can
hide
it.
Yep.