►
From YouTube: 2021-04-09 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
I
am
putting
together
a
proposal
to
talk
at
the
strange
loop
conference
in
the
fall
about
being
a
maintainer
and
open
telemetry.
Do
you
know
this
conference
the
whole.
A
Strange
loop
strange
loop:
it's
the
best
conference.
It's
really
awesome
the
only
bummer
about
it,
it's
in
st
louis!
But
aside
from
that,
it's
a
it's
kind
of
a
it's
a
software
conference,
but
it
brings
together
academics
and
artists
and
researchers
and
practitioners
and
it's
yeah.
It's
a
super
cool
fun
conference.
B
C
C
Yeah
I
had
did
not
rely.
I
mean,
I
know
that
the
the
us
obviously
with
the
drug
companies
here.
A
A
B
B
A
Similarly
yeah,
so
one
of
the
one
of
the
tracks
at
strange
loop
is
sustainable,
open
source.
So
I
thought
it
would
be
a
good
time
to
talk
about
you
know,
being
a
maintainer
on
a
gigantic
multi-vendor
multi-language,
open
source
project.
A
I'm
planning
on
type
my
title,
my
working
title
is
sleeping
with
the
enemy.
B
I'm
going
to
be
giving
a
presentation
internally,
but
not
to
engineers
to
random
people
in
the
japan
office
about
working
with
competitors.
A
It's
very
interesting:
it's
definitely
nothing.
I've
had
to
do
before
so.
B
B
B
D
B
C
A
C
I
I've
gotten
the
the
vibe.
A
A
D
B
C
B
C
The
lattice
thing
came
up
recently:
they
in
their
code.
They
capture,
is
some
events
like
this
starting
code
and
encode
and
which
are
not
you
know,
part
of
this
bag.
But
since
we
pull
in
sort
of.
B
B
B
C
Yeah
that
particular
one
it
just
created
a
lot
of
noise,
noisy
events
in
our
system,
so
I
just
suppressed
it.
B
C
C
C
Oh
yeah,
no,
because
we
kind
of
went
we
kind
of
like
silently
like
dropped,
because
we
had
the
agent
already.
We
sort
of
have
just
replaced
our
old
agent
with
the
new
agent
and
kept
the
configuration
stuff
and
a
lot
of
the
other
stuff.
The
same,
which
has
the
upside
of
getting
a
lot
of
customers
onto
the.
D
B
A
B
So
far,
you're
the
only
one
that
talked
about
it.
So
if
more
users
are
having
trouble,
then
I'll
prioritize
debugging
it
I
think,
but
for
now
I'll
see
if
the
new
version
just
by
itself
helps
if
it
does,
that
would
be
nice
yeah.
I
might.
I
might
try
that
myself
just
on
my
gaming
machine
see
what
happens.
C
B
C
B
C
Because
we
had
some
weird
great
sporadic
breakage
at
the
exact
same
time
and
our
internal
build
on
github
with
the
docker
containers.
Talking
back
to
our
some
networking
between
the
docker
container
and
the
host,
and
I
tried,
I
tried
to
roll
back
to
18
and
that
didn't
fix
it
either
and
eventually
it
was.
We
were
doing
something
weird
with
the
the
doc
we
weren't
using
a
standard
docker
way
to
talk
to
the
host.
C
B
C
C
B
B
D
C
The
one
on
the
sdk
side,
yeah.
B
C
So
a
couple
of
things,
but
in
addition
to
this
I'd
like
to
get
in
this,
this
seems
like
an
important
fix.
B
C
Oh
because
that
I,
because
I
updated
every
single
http
client
yeah.
B
C
C
I
did
end
up
breaking
out
the
that
do
request.
Do
you
do
reused?
Requests
that
worked
out?
Well,
there
were
two
two
that
didn't
really
have
a
request
concept,
so
I
just
parameterized
it
with
void.
C
B
C
Yeah,
I
was
gonna
start
working
on
the
release
notes
today,
but
didn't
get
to
it
so
yeah.
When
I
I
will
try
to
do
a
first
pass
at
least
of
the
release
notes.
I
think
there's
going
to
be
a
vault,
because
it's
a
month
now
and
it's
been
a
fairly
busy
month.
D
C
D
B
B
C
C
Let's
see,
I
think
there
were
a
couple
things
I
was
supposed
to
chat
with
you
about
oh
yeah,
the
oh,
this
one
we
just
chatted
about.
This
was
when.
B
C
Happen
so
I
will
post
in
our
back
channel
so
that
nikita
knows
the
plan
and
then
the
other
thing
is:
what
do
you?
We
were
talking
about
the
change
log
process
that
the
sdk
follows
and
whether
that
would
be
worth
doing
in
the
instrumentation
repo.
If
that's.
B
C
Prs,
what's
your
john,
what's
your
typical
like
how
many
do
you
queue
up
when
you
an
update.
A
I
don't
know
we'll
search
the
the
java
repo
for
changelog
vrs,
I
mean
it's.
Usually
I
probably
do
it
kind
of
weekly,
rather
than
based
on
on
some
number
of
things
that
have
changed.
C
Yeah
yeah
cool
all
right.
So
then.
C
A
Dealt
with
so
it
it's,
it's
not
well.
There's
no
formal
process
that
I
have
defined
for
myself.
B
B
C
C
Yeah,
oh
that
yeah,
the
doc
update
stuff.
We
talked
about
john
shared.
How
is
going
in
the
sdk
repo?
B
B
B
A
A
A
I
have
no
idea
their
docs
are
very
like
their
whole
thing
is
very
out
of
date.
C
C
C
A
D
C
C
B
B
A
C
A
B
D
B
So
one
of
the
interns
decided
to
take
on
this
issue
of
adding
a
builder
for
the
shim,
and
I
was
like
oh,
let's
hold
off
a
bit.
Let
me
try
to
discuss
this
in
the
spec.
If
we
really
need
a
builder
for
this,
because
it's
only
to
be
able
to
decide
these
two
propagators
like
there's
no
other
reason
to
have
configuration
of
the
shim.
B
I
also
wrote
a
new
propagator
for
aws:
it's
not
a
propagator,
it's
a
composite
propagator,
but
one
because
we're
going
to
try
to
start
supporting
more
propagation
like
w3c
and
b3
between
aws
services.
Right
now.
It's
only
that
x,
amazon
thing
so
we're.
Finally,
so
it
extracts
from
all
of
these
formats
and
until
we
can
migrate,
everything
it'll
still
only
inject
x,
amazon,
because
the
downstream
service
might
not
support
it
yet
and
then,
once
it's
all
updated,
then
they'd
start
injecting
the
input
format.
B
Have
you
ever
actually
had
any
thoughts
on
like
how
like
so
we're
a
cloud
provider
we're
not
supposed
to
have
opinions
on
what
formats
people
use.
They
might
be
using
some
old
instrumentation
or
whatnot.
So
I
felt
that
injecting
the
format
you
extracted
is
probably
the
best
behavior
for
a
cloud
provider.
B
That's
a
reasonable
behavior,
but
I
guess
it
could
be
inefficient.
Yeah,
like
we've
had
so
s3
we're
the
ones
who
are
really
like.
Oh
this
trace
headers
so
huge.
What
do
we
do
and
so
they
actually
input
so
they
they're
not
even
using
x,
amazon
they're,
using
basically
the
w3c
binary
format,
because
they
just
wanted
to
keep
it
as
small
as
possible
because
it
goes
through
like
hundreds
of
notes
within
like
any
request.
S3
goes
through
a
bunch
of
storage
servers.
I
guess
so
they
didn't
want
to
propagate
that.
B
C
How
do
you
know,
how
do
you
keep
track
of
what
format
was
extracted.
C
C
A
C
Yeah,
I
was
thinking
I'm
stuck
in
thinking.
Samplers
samplers
can
only
populate
trace
things,
but
yeah
right.
A
Either
one
is
fine
yeah,
that's
on
whether,
like
the
like,
the
tradestate
suggestion
you
had
on
rock
by
the
way
work
great,
the
customer
was
perfect.
They
were
very
happy
to
use
the
trade
state
to
carry
that
information
they
needed.
I
mean
the
nice
thing
about
trace
data.
Is
it's
a
little
bit
fewer
allocations
like
you,
don't
need
to
make
a
context
copy
and
it's
available
to
the
api,
which
should
key
the
key.
Probably
if
you
have
a
propagator,
that's
injected.
A
A
C
Yeah,
I
don't
think
yet
that
the
the
back
end
services,
don't
then
report.
B
C
I
know
that
johannes
from
nurella,
who
john
knows
from
new
relic
is
that
he's
on
the
team
here
that's
doing
that
stuff,
so
he
would
he
would
know.
I
will
ask
him
next
time.
B
C
Yeah
yeah,
so
microsoft
has
been
like
they
were
part
of
the
w3c
thing.
A
B
A
C
So
so
the
reason
it's
different
than
a
multi-propagator
is
because
you
don't
want
to
propagate
out
everything,
and
you
want
to
train.
You
want
to.
D
B
C
A
B
C
B
A
A
We
just
have
a
flag
that
we
could
set
somehow
that
would
turn
them
off.
So,
if
I'm
like,
when
I'm
working
on
my
windows
machine,
I
can
be
able
to
run
clean,
builds
yeah.
The
funny
thing.
The
thing
I
don't
really
understand
is
it's
very
like
so
the
test
fails
and
then
I
run
it
again
and
everything
works
fine.
A
C
Is
there
I
don't
know
we
we
get
that
behavior
a
lot
in
the
instrumentation
repo,
where
if
we
have
two
tests
that
sort
of
depend
on
that
don't
work
well
together
and
the
first
time
you
run
it
test,
a
will
pass
and
test
b
will
fail,
and
then
you
re-run
it
and
test
a
was
cached
now
because
it
succeeded
and
it
will
only
run.
A
C
B
A
Yeah,
if
you
have
a
windows
machine
you
can,
you
know,
get
wsl
on
and
try
it
out
and
also
be
good
to
have
another
test
case.
I
have.
I
don't
have
windows
pro,
so
I
can't
install
docker,
although
nikita
was
claiming
there's
a
way
to
install
docker
on
windows
home,
like
windows
10
like
personal
edition,
but
I
haven't
been
able
to
figure
it
out
like
I
can
install
docker,
but
I
can't
run
it
because
there's
no
virtualization.
A
B
A
C
Tell
splunk
you
need
a
windows
box.
A
A
linux
box,
then
I
would
be
waiting.
Linux
is
nice
yeah,
although
I
did.
I
did
start
sorry,
jason's
recommendation.
I
started.
I
installed
ubuntu
multipass.
If
you
ever
use
this
at
all,
so
it
lets.
You
run
a
non-docker
ubuntu
image
like
using
their
own
virtualization
thing
and
it
keeps
state
unlike
docker.
So
it's
really
nice.
A
It's
a
very
intense
ubuntu,
although
you
have
to
do
some
real
gnarly
hacks
to
get
it
to
have
more
than
one
core
and
more
than
a
gig
of
ram,
but
there
are
posts
about
how
to
do
it,
but
it
seems
kind
of
good
seems
nice.
So
I
was
able
to
test
ubuntu
20
on
that
and
have
it
fail
locally
on
my
mac,
which
is
nice.
B
A
Yeah,
I
take
a
look
at
multi-pass.
It
seems
really
easy
and
a
little
nicer
if
you
can
get
enough
ram
and
cpu
and
enough
cores.
That
was.
That
was
a
funny
thing
about
that:
doordash,
blog
post
I'd
never
seen
the
the
use
of
millicores
I'd
say
I
saw
he
was
using
a
thousand
millicourse
in
this
test
environment.
A
I
was
actually
surprised
that
he's
able
to
get
anything
useful
out
of
a
thousand
millicourse,
because
that's
really
just
a
core
right,
but
I
guess
that's
enough
yeah,
but
I
guess
maybe
you
can
run
enough.
You
can
get
enough
slicing
of
something
across
the
fractional
cores.
I
don't
know
very
strange
way
to
think
about
the
world.
A
I
don't
know
how
I
don't
know
what
a
thousand
millicourse
means
like
that's
just
one
core
on
one
one
cpu
like
you
can't
run
java
effectively.
You
certainly
can't
run
gradle
effectively
on
one
on
single
core.
It's
just
like
your
tech.
Your
bill
is
going
to
run
for
not
our
not
our
build
right,
but
not
even
the
regular
open
songs,
for
it
takes
a
really
really
long
time
on
a
single
core
like
20
25
minutes.
A
B
A
C
Oh
yeah
nikita
was
thinking
about
this
some
more
and
he
had
what
was
his.
C
So
tyler
gave
us
sort
of
the
high
level
of
this,
which
is
that,
when
bite
buddy
needs
to
look
at
the
type
definition
it
it
doesn't
have
any
internal
caching.
So
if
it
has
to
look
up
a
type
definition
again
later,
it
has
to
rebuild
that,
and
I
think
that
can
be
expensive
because,
like
you,
if
you
may
have
you
have
to
parse
the
class
file
to
build
the
type
definition,
because
you
don't
want
to
load
the
actual
class
instead
of
instrumentation,
because
then
it
won't
get
instrumented.
C
And
so
this
is
just
a
pool.
Caching
limited
caching
strategy
on
top
of
that
and
you're
saying
they
used
to
do,
have
two
layers
of
where
first
it
was
the
cache,
the
class
loader.
C
And
then
it
was
oh
yeah.
Here's
the
key.
It
used
to
be
a
two
level
cache
one
to
the
class
loader
and
then
to
the
name,
but
that
made
it
hard
to
cap
consistently
like
at
a
hundred
say.
So
that's
why
it's
combined
here
now.
C
But
then
taylor
mentioned
that
the
particular
problem
that
nikita
was
having
of
the
bite
buddy
not
being
able
to
oh.
So
what
nikita
was
trying
to
do
was
to
load
instrumentation
dynamically
inside
of
that
extension
also,
and
so
that
ins,
the
the
cash
provider
that
this
was
not
finding
his
new
instrumentation
classes
and
what
tyler
thought
was.
It's
likely
that
on
the
child
possible
on
the
child
class
loader
that
nikita
is
building
for
isolation
underneath
the
agent
class
loader
that,
if
you
don't
implement,
get
resource
correctly,
then
get
resources.
D
B
C
It's
a
good
question
I
feel
like
so
with
isolated
class
loaders.
Well,
let's
see
with
the
isolated
class
loader
you
can
delegate
to
the
parent
on
load
click
yeah,
that's
right!
That
would
skip
the
child's
get
resource
on
you,
but
if
you
were
pulling
in.
B
B
C
And
I
think
that
oh
john
made
a
call
out
for
opinions
about
the
g.
What
to
do
about
the
grpc.
B
What
did
I
did
I
respond
on
this
yesterday?
I
forgot.
A
B
C
B
B
B
A
A
B
Yeah
I
mean
since
we're
gen
generating
the
produce
ourselves.
If
we
shade,
like
everything,
would
get
shaded,
so
it
would
work
normally.
You
can't
change
your
pc
because
the
products
tend
to
not
be
shaded
and
published
somewhere
and
then
they're
not
compatible,
but
if
you're
shading
the
produce
together
with
your
piece,
it's
possible.
A
B
A
B
C
What
about
a
have
the
default
be
like?
Have
it
have
the
shading
at
the
exporter
level,
so
that.
C
B
A
B
C
B
C
Well,
the
so
it
sounds
pretty
clear
that
I
mean
if
we
did
pick
one
if
we
did
pick
one.
We
would
pick
this.
B
C
For
kind
of
a
similar
reason
that
we
had
with
the
with
that
security
issue,
also
yep.
B
But
so,
for
example,
if
my
app
is
using
jrpgneti
and
open
templates
using
jrpc,
I
didn't
exclude.
That
means
that
I
have
this
shaded
netty,
as
well
as
the
normality
in
my
app
okay.
It's
not
the
end
of
the
world,
no
yeah.
So
that's
the
and
then
there's
the
other
issue
where,
if
I'm
using
jrpc
1.37
without
the
bomb
in
my
app
and
then
I
depend
on
open,
telemetry
depends
on
1.36.
C
B
B
C
C
B
B
A
C
A
C
What
do
you
I
mean
doing
it
at
the
exporter?
Lit
level
would
give
would
prevent
the
bad
failure
mode
and
the
only
the
failure
mode
then
is
basically.
Okay.
We've
pulled
in
way
too
much,
and
if
somebody
cares
about
that,
they
can
then
go
and
use
they're
talking.
B
I
mean
I
like
I
so
maybe
the
way
I
would
put
it.
Maybe
you
can
double
check
whether
it's
a
reasonable
statement
or
not,
if
you're
adding
a
dependency
anyways
as
long
as
our
docs
say,
add
this
dependency
as
well.
At
the
same
time,
you
add
the
two
dependencies
and
you're
good
to
go
right.
Is
this
not
a
normal
thing
to
expect
of
users?
B
C
D
B
B
C
B
C
Did
we
have
any
pr's
that
I
needed
to
look
at.
B
C
Yeah,
should
we
go
ahead
and
ask
for
because
I.
D
C
Always
ask
for
repros,
like
that's
that's
one
of
the
the
contracts
so
that
one
of
the
things
I
like
about
open
source
is
like
it's
like
hey
you're,
getting
this
for
free.
You
want
to.
You
want
help
with
your
help,
help
us
help
you.
B
We
had
this,
so
the
last
comment
mentioned
would
span
that
does
scare
me
a
bit
because
I
do
wonder
whether
what
the
agent
does
to
kotlin
methods
like
what.
If
it's
a
suspending
function
or
I
guess
we
would
still
find
it
instrumented
and
that
would
probably
break
like
that
seems
guaranteed
to
break
something.
C
B
Yeah,
I
mean,
I
don't
know,
maybe
kotlin
just
doesn't
carry
over
annotations
or
something
but
like
I
can
imagine
kotlin
splitting
it
up
into
two
functions.
One
of
them
has
the
with
maybe
both
of
them.
I
will
do
it
span
and
then
we
find
both
of
these
the
start
and
the
end
of
the
split
suspending
function
instrumented
and
that
is
not
going
to
work.
I'm
sure.
B
B
So
I
haven't
found
a
good
way.
If
you
looked
at
my
strict
contact
storage,
I
walk
the
stack
to
see
patterns
of
whether
it
looks
like
a
suspended
function
or
not.
Yeah.
Okay,
that's
all!
I
know
there
are
my
ideally
there's
a
better
way,
but
at
least
I
couldn't
find
it
so
it
isn't
then
just
a
jvm
function
and
I
don't
know
how
they
would
communicate.
C
And
possibly.
C
Yeah,
it's
interesting
how
like
in
our
issues,
I
feel
like
skew
so
much
more
towards
kotlin
and
grawl.
That
then
really
exists
in
the
real
world.
C
B
B
C
B
C
Good
deal
all
right
well,
have
a
good
have
a
good
friday
and
yeah,
so
we'll
keep
in
touch
about
the.
But,
yes,
I
will
work
on
the
release,
notes
and
then
coordinate,
hitting
the
button
with
nikita
cool.