►
From YouTube: 2021-08-06 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
Honorary
so
today
I
don't
know
if
you
see,
I
would
picked
you
on
justin's
pr
about
the
nullable
annotations.
A
Oh,
it
was
just
a
few
minutes
ago.
I
was
curious
if
I
was
right
or
wrong
about
my
reading
of.
I
know
how
checker
framework
works,
which
is
confusing
for
nullable
arrays.
B
B
C
B
A
B
A
B
B
A
Hit
the
mark
of
especially
for
being
open
sort
like
a
contributor
friendly.
A
A
And
ludmilla
had
two
open
questions
and
mateish
had
one
open
question.
So
one
was
whether
to
do
the
set
type
or
the
auto.
A
A
A
Oh
yeah
and
she
was
thinking
like
if
there's
multiples,
different
ones,
then
maybe
just
make
them
generic.
Like
I
mean
we
don't
because
it's
something
pretty
unique
at
that
point
like
I
don't
think
we
have
any
examples
in
our
instrumentation.
That
applies
to
different
semantic
conventions
to
one
span.
A
Sorry
to
say:
what's
your.
A
B
A
Cool
that
makes
sense
to
me,
so
I
will.
A
And
then
oh
yeah!
This
is
why
I
pinged
justin
on
that
with
span
annotation
pr,
because
ludmilla
mentioned
that
the
while
trying
to
apply
this
across
the
repo.
It
was
a
challenging
because
the
with
span
annotation
still
uses
tracers,
and
so
she
was
hoping
to
only
instrume
only
modify
instrumenters
and
not
tracers.
So
so
I
think
this
this
kind
of
solves
itself
once
that
pr
is
merged.
A
A
And
then
materia
she's.
A
Making
sure
that
the
implementation
is
bridge
friendly,
so
bridging
the
context
keys,
I
think
the
current
pr
implementation
is
not
so
just
talk
a
little
about
what
that
means,
which
I
didn't
quite
follow
because
I
forgot
I.
B
A
B
B
A
B
A
Instrumentation
by
default
and
so
yeah,
so
hopefully
that
won't
be
a
surprise
when
that
comes
along
eventually.
A
Moment
somebody
doing
the
work.
Oh
I
see
here
are
the
people
that
aren't.
D
Doing
the
work,
I
guess
yeah
I
mean
everyone
seems
to
agree
that
that's
a
good
idea
right,
that
the
core
and
any
stuff
should
just
be
disabled
by
default.
There's
only
one
thing
that
so
it's
it's
the
one
who's
resistant
to
it.
B
D
A
For
the
rap
pack-
and
that's
one
of
the
blockers,
john-
is
that
we
have
to
instrument
more
stuff,
so
we'll
have
to
instrument
rap
pack
as
a
server.
D
D
B
B
D
D
B
I'd
say
it's
the
same
when
you're
using
I
mean
I
guess
some
servers
might
tell
you
to
configure,
but
usually
like
with
http
2.
You
just
have
one
connection
and
it's
like
pipelining
is
just
part
of
the
spec.
It's
not
a
feature,
so
we
would
generally
just
use
the
word
http.
B
B
B
D
B
B
C
B
B
A
C
B
B
Red
pack,
hd
decline,
just
give
me
some
headaches
because
it's
very
hard
to
write
tests
for
it
rat
pack,
hp
client.
You
can't
use
standalone
like
it's,
not
meant
to
be
a
standalone.
It's
supposed
to
be
used
in
a
rackpack
server,
so
they
have
some
test
harness
to
try
to
allow
you
to
at
least
test
for
it,
but
the
test
harness
doesn't
reproduce
the
wrap
back
server
that
well
and
so
you've
used
rack
pack
before
john.
D
A
D
Yeah,
I
never.
I
was
never
able
to
wrap
my
head
around
rat
packs,
currency
model
or
anything.
It
was
super.
D
To
me
yeah
my
big
takeaway.
After
maintaining
that
thing
for
a
year
and
a
half
was
like,
if
I
could
write
this
again,
I
would
definitely
not
use
rat
pack
yeah
it's
a
kind
of
a
crazy
like
I
even
watched
all
the
videos
when
they
were
describing
how
their
their
future
system
works
and
everything,
and
I
just
my-
I
couldn't
wrap
my
head
around
it.
It's
very.
B
D
B
D
C
D
But
I
will
say
also
new
relics:
rat
pack,
instrumentation
didn't
work
all
that.
Well,
so
maybe
we
don't
get
tyler
to
do
it.
I
mean
it
kind
of
worked
for
the
most
part,
but
there
were
little
gaps
and
weird
holes
and
things
would
like.
I
think
there
was
some
real
problems
with
being
able
to
detect
when
a
request
failed
before
it
hit
the
user
land
code
and
capturing
like
the
status
codes
and
things
like
that
was
kind
of
a
mess.
B
D
D
The
api,
well,
he
wrote
when
he
wrote
it
while
he
was
maintaining
the
api
gateway,
but
it
was
only
just
an
internal
extension
that
I
don't
think
was
perfect.
It
wasn't
officially
published
anywhere.
D
B
C
B
D
Yeah,
that
was
my
other
big
takeaway.
After
maintaining
that
api
gateway
was,
I
probably
actually
wouldn't
use
async
on
as
an
a
server
is
what
I
ended
up
deciding.
It
was
just
super
hard
to
reason
about
scaling
and
when
you
know,
you're
going
to
pin
a
thread
to
a
connection,
it
was
very
it's
very,
very
easy
and
easy
to
reason
about
scaling
and
async.
D
It's
very
difficult
to
figure
out
how
you
need
to
be
able
to
scale
your
systems,
especially
when
you
have
a
when
you
have
a
very
heterogeneous
request.
Structure
like
an
api
gateway
does
because
we
were
basically
serving
apis
from
all
over
the
company,
and
some
of
them
were
very
slow,
and
some
of
them
were
very
fast,
but
figuring
out
so
figuring
out
how
to
scale
that
particular
system
and
be
have
it
be
resilient
to
bursts
in
fast
or
slow
back
ends
was
very,
very
difficult.
A
A
D
D
D
Like
we
would
have
old,
you
don't
have
to
convince
me:
we'd
have
old,
new,
relic
rails
back
end
that
was
still
out
there.
That
would
be
like
yeah.
We
have
timeouts
of
a
minute
and
a
half,
because
sometimes
the
queries
take
a
super
long
time,
and
then
there
would
be
other
services
that
had
timeouts
of
like
four
seconds
or
half
a
second,
and
being
I
mean
anyway,
it
was
yeah.
It
was
very
tricky
to.
A
Manage
anyway,
but
I
do
feel
like
like
I
mean,
and
yes
it's
all
for
me
about
that
reasoning
of
like
I
would
like
to
be
able
to
take
a
stack
trait,
let
you
know,
thread,
dump
and
see
what's
running
in
my
system,
but
that's
where
I
feel
like
you
know
someday.
I
feel
like
we'll
get
there
with
async
and
reactive
stuff,
but
we
are
not
there
today.
D
So
I
got
a
really
good
comment
from
laurie
on
my
pr
that
I
put
in
to
document
this,
that
basically
our
library
instrumentation
for
async
doesn't
work
because
of
the
way
rat
pack
or
sorry.
Ok,
http
does
is
dispatching.
B
D
So
so
the
way
that
okay
does
its
so
you
can
plug
in
a
dispatcher
which
wraps
an
executive
service,
but
the
way
that
it
does
its
dispatching
is
that
it
will
collect
up
all
of
the
outstanding
requests
and
try
to
dispatch
them
and
they
those
may
have
come
from
other
threads.
D
So
by
the
time
our
rapper,
the
task
wrapping
rapper,
comes
along
and
can
wrap
the
actual
runnable
that
is
being
dispatched.
It
may
be
running
on
a
different
thread
than
when
it
started
because
of
their
their
huge
dispatching
system
that
they've
built
so
library
instrumentation.
We
don't
have
a
way
right
now
to
actually
propagate
the
context
properly
into
the
request
for
async.
Okay,
it
should
be
into
the
request
or
into
the
callback.
D
D
No,
it's
that's.
The
causality
is
turned
off
for
the
library
instrumentation
for
the
causality
with
callback
is
turned
off
right,
yeah,
yeah,
no
without
callback,
it
works
fine,
but
as
soon
as
you
have
callbacks
that
it
does
not
like
the
dispatching
system
does
not
preserve
the
thread.
The
calling
thread
so
dispatching's
just
different.
A
Okay,
yeah,
I
did
not
realize
that
I
thought
it
was
only
the
call
back
the
that
the
context
didn't
get
propagated
to
the
callback
so
that.
D
D
D
A
A
D
D
The
upshot
is
that
we
don't
have
a
reliable
way
with
library
instrumentation
for
okhttp
in
the
async
case
in
the
enq
case
to
actually
propagate
context.
D
I'm
sure
we
can
do
it
with
the
bytecode
instrumentation,
because
you
can
hook
directly
into
the
the
actual
executive
service
and
when
that
get
that
specifically
gets
called,
then
you
can
grab
the
grab
the
scope
and
do
the
things
you
need
to
do,
but
it
doesn't
doesn't
work
with
the
library
instrumentation,
which
is
a
way
to
brave.
B
B
I
thought
it
generally
works
with
so
brave
documents.
These
two
patterns
implemented
the
second
one,
because
I
thought
that
still
works,
but
it
actually
recommends
not
using
the
location,
client
interface,
please
call
that
factory
and
so,
instead
of
interceptors,
it
just
wraps
the
call
out
factory
just
like,
as
if
it
was
a
method
to
call
and
it
instruments
the
callbacks
and
whatnot
said
that,
of
course,
propagates.
Fine,
because
all
the
threading
is
everything
happens
on
the
collar
thread
right
yeah,.
D
If
you
do
like,
if
you
do
your
own
async,
like
you,
have
your
own
executive
service,
where
you,
where
you
make
synchronous,
okay,
http
calls,
then
ever
you
can
make
everything
work,
just
fine
as
well,
but
then
you're
managing
it
rather
than
using
the
callback
based
one.
It's
definitely
more
of
a
pain
like
the
callback.
Api
is
really
is
very
handy.
D
D
I
mean
unless,
unless
you
have
a
different,
a
weird
different
case,
where
you're
trying
to
do
a
whole
bunch
of
background
processing
and
hi,
I
just
can't
imagine
an
android
app
where
you'd
want
to
have
a
whole
bunch
like
very
high
throughput,
http
client
access,
except
if
you're,
like
loading
tons
of
images
like
on
a
google
map,
or
something
like
that.
I
suppose
that
could
happen
anyway.
D
I
don't
think
there's
a
good
solution
to
it
at
the
moment,
but
we
should
probably
document
something
in
the
end
that
just
says
like
hey.
This
is
not
going
to
work
very
well
for
async
with
the
library
instrumentation
because
it
certainly
looks
like
it
will
work
and
when
you
do
low
volume
testing
on
it,
it
works
just
perfectly
it's
not
until
you
get
to
high
contention
the
things
that
okay
tries
to
be
super
clever
and
messes
up
the
parenting.
D
A
Out
but
without
the
dispatcher.
D
D
It
will
be
consistently
misparent
or
no
parent.
It
will
have
no
parent,
it's
really
surprising
yeah,
it's
a
bummer.
I
don't
know,
there's
no
good
answer
here
for
library
instrumentation,
I'm
not
sure
what
to
do
about
it,
but
at
least
we
should
document
it
and
I'll.
Add
that
to
that
pr
and
just
say
hey,
this
is
probably
not
going
to
work
for
you
under
high
throughput
scenarios.
Multi-Threaded
high-throughput
scenarios.
A
And
what
were
you
saying
about
this
option?
On
iraq,
I
mean.
B
D
A
B
B
Missing
the
main
reason
I
sort
of
preferred
using
http
client
was,
I
was
sort
of
imagining
using
this
in
the
kubernetes
client
instrumentation,
which
doesn't
use
called
our
factory.
They
just
use
ok,
http
client,
but
never
actually
arrived
to
doing
that
either
way.
It
looks
like
we
should
just
follow
this
pattern
of
providing
the
tracing
call
factory.
B
Well,
so
call
factory
is
the
http
interface.
Oh
okay,
http
client
implements
call
factory.
A
B
B
C
B
D
C
D
B
C
B
Class
that
wraps
a
bunch
of
logic,
so
it's
not
as
big
a
deal
probably
but
anyways
brave
doesn't
seem
to
use
that
pattern.
So
I
can
refer
to
brave.
Also
then.
C
D
Yeah,
that's
a
that's
a
bigger
thing,
I'll
I'll.
My
pr
will
just
document
the
the
current
the
current
limitations,
but
we
should
probably
create
an
issue
to
track,
and
I,
if
I
have
time
I'll
pick
that
up
and
try
to
see
what
I
mean,
I
could
see
the
pattern
here.
B
D
D
A
B
D
D
B
B
It's
a
good
find,
though,
like
I
didn't,
realize
the
importance
of
the
call
factory.
So
now
that
I
know
we
can
prioritize
that.
C
A
Yeah
yeah.
A
And
just
they're
just
getting
ron
is
that
developer
working
on
it
and
just
giving
an
update
kind
of
the
they're
just
getting
going
but
great
to
have
that
going
forward
in
parallel
to
everything
else.
A
John
had
a
couple
of
pointers
and
dumped
in
the
links
to
splunk's
log
stuff
and
the
old,
the
old
log4j
instrumentation
that
we
had
for
inspiration.
A
Oh
yeah
nikita
what
was
nikita.
Oh
yeah,
he
was
looking
for
comments
on
the
url
based
sampler
and
we
talked
a
little
bit
about
http
url
versus
route,
in
particular,
whether
we're
like
work
that
we're
currently
capturing
http
url
everywhere,
but
we
may
we
may
or
may
not
change
that
and
nikita
was
thinking
we
wouldn't
ever
capture
on
target
sun,
because
it's
like
very
low
level.
It's
like
defined
as,
what's
in
the
header,
the
the
http,
whatever
first
line
of
the
http
request,
but
I
mentioned
that
I
mean
it's.
A
Oh
to
get
the
path
right
yeah,
so
I
think
he
was
actually
in
favor
of
in
that
case
capturing
I
mean
the
path
instead
of
building
the
url
and
then
because
it
sounds
like
in
here
he's
taking
the
url
apart
to
get
the
path
portion,
yep.
D
A
B
D
D
D
Or
something
like
that,
so
that's
a
really
quick
and
my
comment
further
down
where
I
answered
all
his
like,
I
gave
my
answers
to
his
four
questions.
Was
that
looking
at
kind
is
a
very
cheap
check,
because
it's
just
a
getter
on
the
readable
span.
D
So
if
you,
if
you
do,
want
to
restrict
unkind,
it's
very
easy
to
it's
a
very
it's,
a
very
lightweight
check
to
eliminate
like
internal
spans
and
not
even
bother
to
look
at
them.
For
example,.
A
This
sampler
there's
no
like
readable
spam
thing
right.
D
D
A
D
Not
on
spam
processor
right
in
a
read,
read
a
box
band.
Readable
span
has,
I
believe,
has
access
to
the
attributes.
B
D
D
A
A
Yeah
yeah,
that's
that's
my
use
case
here
and
that's
why
this
is
just
a
draft
pr
to
show
that
this
is
gonna,
be
painful.
If
we
want
to
implement
this,
but
we
have
to
yeah
get
here.
A
I
was
thinking
of
potentially
just
since
this
is
running
in
the
agent
anyway,
I
know,
but
then
there's
the
whole
concurrency
stuff.
So
I
don't
know.
A
I
want
to
so
we
have
this
concept
of
operation
name
and
in
the
all
like
it's
supposed
to
be
it's
essentially
the
span
name
of
the
request,
but
it's
supposed
to
be
stamped
on
the
like
depend.
The
the
client
telemetry,
the
internal
and
client
spans
also
as
opposed
to
doing
a
joint.
B
B
A
Oh,
it
was
justin,
I
gave
it
the
one,
the
lone
thumbs
up.
A
This
is
essentially,
I
mean,
would
solve
some
of
them
have
another
use
case
like
this,
where,
like
somebody
wants
to
stamp
like
a
tenant
id
onto
all
their
spans
and
but
it's
dependent
on,
like
the
request
right.
When
you
get
the
request
in
that's
where
you
know
the
user
who
the
user
is
and
so
yeah,
you
can
put
that
in
the
context
and
then
write
a
span
processor
to
take
that
out
and
stamp
that
on.
A
So,
and
actually
this
one
wouldn't
be
so
bad,
except
that
we
on
the
on
our
operation
name,
they
really
want
the
the
http
method
in
front
of
the
span
name.
B
A
And
the
only
reason,
just
explaining
the
only
reason
I
even
know
what
the
what's
on
that
readable
span
or
not.
A
Yeah
yeah,
I
really
I
I'm
very
hopeful
that
I'm
worried
that
I
mean
this
instrumentary
api
gives
us
a
lot
of
really
nice
hooks
like
that.
Well,
for
the
auto
instrumentation,
it.
B
B
B
B
D
B
D
B
D
Yeah,
you
should
it's
going
to
help
your
stock
price
trask,
so
you
should
convince
them.
Your
friends
over
at
github.
A
D
Definitely
when
I
would
run
stuff
at
info,
it
would
look
like
the
armeria
server
wouldn't
properly
start
up
a
lot,
but
I
haven't
looked
at
this
particular
case
to
see
if
that's
still
what's
going
on,
and
if
it
is,
I
still
have
no
idea
why
or
how
to
how
to
address
it
anyway.
Maybe
the
recharge
plug-in
is
the
simple
answer,
and
then
we've
got
it.
We
sweep
it
under
the
rug
and
forget
about
it.
B
A
A
B
D
D
D
I
only
I
kind
of
slapped
it
together
and
now
it's
going
to
sit
there
in
my
github
or
my
personal
github
and
never
be
looked
at
again.
You
mean
a
thought.
Honorable
solution
to
gradle
enterprise.
A
D
D
Make
it
a
you
know
a
splunk
dashboard,
that's
like
we
could
have
it:
output
using
the
logging
exporter
and
then
just
put
through
that
into
like
a
github,
the
github
pages,
like
the
way
jason
did
did
stuff
and
then
right.
D
We're
good
we
joke
about
it,
but
that's
actually
not
that
far-fetched.
There's
not
that
many
spans.
It
wouldn't
be
that
big
a
deal
to
just
suck
that
into
a
browser
and
read
through
it
and
do
something
with
it.
I
don't
think
it'd
be
that
big
a
deal
for
someone
who
knew
javascript,
I
mean
it.
Would
I
guess
it
would
be.
I
mean
because
I
know
that,
like
the
java
open,
telemetry
java
generated
around
1200
spans
like
1200
lines
to
parse
through
that's,
not
very
much
data
right,
you
can
load
that
in
the
browser.
D
B
B
Jason
ever
tried,
like
I
always
I'm,
still
interested
in
having
the
internet
benchmark
stuff
as
part
of
the
normal
build
just
so
we
can
apply
our
spotless
conventions
and
whatnot
like
we
have
a
lot
of
java
code
there.
Now
that's
just
not
checked,
I
mean,
of
course
we
shouldn't
do
it
if
it
makes
it
too
annoying,
but
like
in
since
intellij,
you
can
select
the
modules
it
loads
and
if
that
module
is
just
completely
independent
like
that
and
do
anything,
I
think
it
would
work
well
with
all
the
other
modules
deselected.