►
From YouTube: 2021-06-02 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
B
A
C
Evening,
let's
get
that
camera
bed
head
this
morning,
nikita
no
camera.
A
C
Remember
I
was
going
to
ask
you
nikita,
so
we
already
have
handler
spans,
and
so
I
guess
the
question
should.
Are
you
thinking
that?
Maybe
we
should
hide
those
behind
like
a
flag
since
they're,
not
we
don't
have
a
a
spec
for
them.
A
My
first,
what
do
we
already
have
for
hangers
plants?
I
mean
we
create
them,
but
we
don't
treat
them
any
special
way,
comparing
to
other
internal
spawns
or
do
we.
C
Oh,
I
see
what
you
mean
about
nesting
rules,
specifically
around
nesting
yeah
yeah.
A
A
C
Up
these
two
issues,
so
this
issue
here
is
about
that
about
the
nesting.
This
one
is
just
about
what
is
our
do
we
have?
Can
we
agree
on
a
definition
of
handler
span
so
that
we
can
apply
that
consistently
to
the
different
web
frameworks.
C
Play
we,
in
addition
to
you,
know
that
we
capture
handler
spans
some
way
to
be
able
to
communicate.
A
C
C
C
Like
I,
I
kind
of
pulled
out
a
couple
of
examples
of
like
this
one
in
particular
the
webs
flex
handler
span.
Oh
honorag,
you
know
that
you
had
pointed
me
to
the
brave
webflux
interceptor
yeah.
So
what
they're
doing
that's,
because
they
don't
have
netty
instrumentation,
which
makes
sense
that's
kind
of
what
we
would
potentially
like
and
ditch
the
netty
instrumentation,
but
we
what
we're
doing
is
different.
C
We're
capturing
this
handler
span
thing,
but
it's
kind
of
weird,
because
current
today,
the
way
we're
capturing
the
handler
span
is
really
at
that
same
very
high
level.
So
it's
really
the
whole
pretty
much
the
whole
server
request
anyway.
So
I'm
not
really
sure
what
the
point
of
the
webflux
handler
span
is.
A
D
C
Yeah
and
they
inherited
it
from
net
from
new
relic,
I
mean.
I
know
these
are
kind
of
popular
things
to
show
to
users
and
I've.
I've
actually
had
people
ask
us
on
our
product
before
we
had
these
like,
but
new
relic
has
this?
Why
don't
you
and
I'm
like?
Well,
I
tried
to
get
well.
What
value
are
you
trying
to
get
out
of
this
detail.
D
Yeah
I
mean
looking
like
because
they're
not
in
the
open,
sound
respect
if
we
were
to
make
our
instrumentation
from
scratch
just
based
on
the
spec,
these
wouldn't
exist,
and
so,
but
I
think
they
exist
only
because
this
is
for
from
some
older
instrumentation.
So
either
we
get
rid
of
handler's
fans
or
we
have
respect.
I
think
those
are
the
two
options
in.
C
D
Yeah
I
mean
I
personally
wouldn't
use
handles
pins,
but
if
people
like
them,
I'm
sure
we
should
have
them,
but
that
probably
means
that
we
need
to
often
respect,
maybe
get
it
more
formalized
because
otherwise,
like
if
class
on
python,
does
something
totally
different
than
spring.
On
java,
I
mean
even
though
they're
both
web
frameworks,
they
should
give
similar
spans.
D
A
And
I'm
not
sure
this
about
semantic
convention.
What
is
this
is
certainly
about
those
recommendations:
how
to
write
instrumentations
if
you're,
if
you
remember
that
project
of
tests
which
is
not
flying
because
right
now,
but
there
will
be,
I
think,
the
ideal
place
to
describe
how
to
handle
web
frameworks
specifically
as
opposed
to
because
for
me
for
me,
this
is
all
about
modeling
how
we
we
create
span
where
at
at
what
points
in
like
code
path
we
create
spans
and
for
what
purpose,
how
to
handle
next
client
spans
should
be
there.
A
C
You're
thinking
of
that
instrumentation
guidelines
as
sort
of
a
second
layer
below
the
specs,
not
quite
as
formal
as.
A
A
A
A
C
So
do
we
just
table
this
for
now
until
that
work
stream
gets
going?
While
we
I
mean
I'm,
I'm
fine.
A
C
Okay,
is
there
are
there
meeting,
are?
Are
there
meetings
about
this.
C
Okay
yeah,
this
is
we've
got
more
important
things
to
do.
I
just
kind
of
wanted.
It
was
something
I
had
been
has
come
up
from
time
to
time.
D
C
C
Understanding,
I
put
one
other
just
to
get
your
feelings,
because
so
far
we've
got
one
plus
one.
The
one
plus
one
we're
evenly
split.
D
C
C
C
Oh
right,
right,
okay,
yeah.
D
C
C
C
Okay,
yeah,
I
I
don't
have.
I
am
definitely
leaving
I'm
happy
for
you
to
make.
B
D
C
C
C
D
C
C
C
Cool,
let's
see
where
we're
at
on
our
on
our
instrumenter
instrumental
countdown
11
out
of
99.
all
right.
B
B
C
Actually,
we
have
more,
those
are
just
the
ones.
Those
are,
those
are
the
ones
that
need
instrumenters.
There's.
C
Oh
yeah,
there
was
an
intellij
update.
Today
I
was
oh
man.
I
set
a
timer
when
I
was.
I
don't
think
I
remembered
to
stop.
No,
I
didn't
stop
it
as
an
hour
and
12
minutes
ago
that
I
updated
intellij
and
I
was
going
to
see
how
long
it
took
to
do
the
indexing,
but
I
think
it
only
took
like
10
minutes
or
so,
which
was
a
lot
better
than
the
the
two.
D
A
C
A
Yeah
yeah
and
they're
all
different
versions.
Okay,
that's
a
good
point.
I
would
think
about
that
by
the
way
it's
interesting.
I
have
put
two
more
topics
to
quickly
discuss
into
agenda
by
the
way.
A
So
about
extension,
I
got
the
first
feedback
from
the
end
users
and
that
feedback
is
is
mixed
from
one
point
of
view.
The
developers
of
extensions
are
quite
happy.
The
developer
experience
much
easier.
D
So
experience
so
like
I
guess,
stack
driver.
For
example,
the
google
cloud
trace
they've
always
said
they're
not
going
to
make
a
distro
and
prefer
this
mechanism.
So
I
guess
it's
sort
of
at
some
point.
Someone
is
in
control
of
that
user
experience.
If
they
want
to
use
one
jar,
they
can
do
a
distro.
If
they
only
use
two
jars,
they
can
use
the
extension
well.
A
Yes,
so
what
I'm
currently
thinking
in
the
context
of
that
client
of
ours
is
that
maybe
now
that
we
have
an
extension
mechanism,
we
can
somehow
provide
simpler,
repackaging
option
as
well,
I'm
not
sure
about
specific
technical
details,
but
some
way
to
reuse
that
extension
mechanism,
but
still
have
just
one
jar
file
for
a
for
for
everything.
A
Feedback
and-
and
my
second
thought
was
I,
like
that-
yeah
okay,
cool
last
one
good
and
the
second
topic
is
for
the
the
very
same
same
client.
They
have
concerns
that
how
we
allocate
the
a
lot
of
memory
we
meaning
java
agent
and
well.
A
What
does
a
lot
mean
is
very
specific,
of
course,
but
in
their
particular
case,
when
they
have
like
a
high
load
on
their
application
with
agent
and
without
agent
with
agent
we
memory
allocation
was
like
around
two
and
a
half
times
higher
than
without
agent,
so
they
have
pretty
efficient
application
and
when
we
add
agent,
it's
like
two
and
a
half
times
more
memory
allocations.
Judging
from
gc
logs,
which
well
I
can
totally
understand,
it
may
be
a
lot
for
them.
A
A
C
Do
you
I
know
if
this
the
this
20
kilobyte
is?
Does
it
include
end
to
end
over
otlp?
Yes,
yes,
it's
including
exporting
yeah.
Do
you
happen
to
have
the
ones
that
I'd
be
really
interested
in,
are
with
no
exporter
and
then
also
with
sampling
percentage
set
to
zero?
C
I'd
be
really
curious.
Those
I
can
do.
A
That
what
I
so
let
me
finish
and
then
I
will
talk
about
plan
so,
but
when
I
do
micro
benchmarks
in
hdk,
then
I
have
burst
pan
allocations
between
5
to
10
kilobytes,
including
exporting,
which
may
be
still
a
lot,
but
still
like
seems
like
instrumentations,
introduce
more
than
sdk
in
this,
but
for
now
so
I
will
try
to
understand.
What's
what's
going
on
in
agent
and
yes,
so
I
tried
to
run
just
without
exporting
all
instrumentation,
but
without
exporters.
A
A
A
Yes,
and
and
yeah
so
far,
I
don't
understand
where
this
memory
comes
from.
I
think
it's
too
much
again.
John,
had
very
good
suggestion.
Yesterday,
how
can
we
estimate
what
is
too
much?
I
mean
he
said
that
we
have.
We
have
lower
bound,
it's
serialized
or
tlp.
A
That's
the
data
that
we
produce
and
that's
effectively
lower,
lower
bound
for
allocation.
We
cannot
allocate
less
than
that,
so
I
will
have
to
measure
how
much
in
the
like,
in
example,
application.
Do
we
have
on
the
wire
that
give
gives
up
some
new
ideas
but
yeah
so
far,
all
those
sampling
memory
profilers,
which
may
which
measure
it
like
kill
up
sampling.
A
C
So
no
no
tools
to
get
like
allocation
profiles
of
like
hot
allocation
hotspots
of
like
I've
used.
I
don't
remember
what
it
I
thought
it
was.
I
thought
jfr
had
this
option,
but
maybe
it
was
something
else
that
it
would
sample
some
small
percentage
of
your
allocations
of
your
new
objects
and
it
would
capture
the
stack
traces.
A
A
Apart
from
whenever
I
take
that
memory
allocation
profile
it
it,
it
shows
to
me
that
batch
ban
processor
worker
thread
allocate
the
majority
of
its
memory
while
taking
those
get
get
bytes
from
trace
id
or
get
fan
id
as
by
which
should
be
like,
like
like
maximum
80
80
bytes
per
span
of
40
bytes
per
span,
but
I
see
like,
but
I
see
in
jfr
memory
profile
like
30
to
50
times
more,
so
something
is
fishy
somewhere.
A
C
So
you
said
spring
pet
clinic
is
what
you're
using.
D
D
C
A
And
strangely
enough,
they
produce
similar
number
per
spa,
which
I
was
surprised
because
well,
jdbc
instrumentation
has
those
sql
queries
inside.
So
I
assumed
at
the
beginning
that,
like
the
size
of
attributes
can
matter,
but
I'm
not
so
sure
anymore,
so
something
else
is
going
on,
but
well
I
will
rewrite
them
anyway.
Yep.
A
C
Yeah,
I
need
to
dust
off
my
benchmarking
up,
so
I'll
try
to
do
that
also
and
start
taking
a
peek.
It
won't
include
otlp
because
it'll
be
our
exporter,
but
that
could
be
interesting
also.
C
Did
yeah
that's
why
I
was
bugging
tyler
last
week
for
the
the
thread
dump
for
the
deadlock
yeah.
Did
you
see
that
I
don't
think
I
saw
the
bugging
last
week?
No
because
it
was
on
it
was
on.
I
just
followed
up
on
that
super
old
thread.
You
got
it
our
slack
channel,
so
he
gave
he
was
able
to
get
the
post
the
deadlock,
the
thread
dump.
C
And
it
looked
to
me
like,
first
of
all,
it
was
this
really
amazing
deadlock
cycle
of
five
threads.
It
was
the
minimum
like
I've.
Never
I
don't
think
I've
ever
seen
a
minimum
deadlock
cycle
of
five
threads,
but
one
of
the
components
of
the
deadlock
was
a
lock
that
they
take
in
their
agent
class
loader,
which
we
don't
take
and
they
have
it
because
they
have
those
multiple
agent
class
loaders
for
those
different
folders
the
inst,
and
they
have
a
couple
others
and
nikita.
C
Actually,
if
you
didn't
see
that
I
did
wanna,
I
was,
as
we
start
to
have
multiple
class
loaders.
I
don't
think
we
need
that
lock
still
even
with
our
multiple
class
loaders,
because
we're
not
loading
the
same
class
see
the
problem
is
that
they
were
loading
this
they
needed
it
because
they
were
loading,
the
same
classes
from
different
class
loaders,
as
opposed
to
what
nikita
is
done
with
the
extension
class
loader
is
delegating
to
the
parent.
C
I'm
not
really
clear
why
they
didn't
do
that,
but
anyway,
so
they
have
a
lock
in
there,
basically
on
class
loading,
and
so
that
was
one
component
of
it.
There
was
also
components
of
the
caffeine
right
under
had
a
lock
and
inside
of
it's.
It
was
loading,
more
classes
right,
which.
A
D
Yeah,
if
you're
convinced
I'm
convinced
it's
sort
of
it's
been
a
long
time,
but
I
remember
not
being
able
to
figure
out
why
I
would
deadlock,
that's
why
you
were
discussing
it
like
if
it
loads
classes,
it's
supposed
to
be
okay
right!
That's
why
I
was
always
confused,
but
that
sounds
about
right.
If
you're
loading
classes
and
have
a
separate
look
that
then
perfect.
D
D
D
C
C
D
D
C
Unless
it
always
picks
like
the
lowest,
I
don't
know,
but
yes
sounds
good.
Oh
yes,
I
said
this
was
for
you
nikita.
A
D
C
This
cancellation
thing
is
weird,
not
not
halo's
one,
but
the
the
spring
web
flex
I
mean.
First
of
all,
I
was
so
thankful
that
laurie
tracked
that
down.
I'm
wondering,
though,
like
because
on
cancellation.
D
D
D
D
D
D
So
it
might
complete
or
but
there's
probably
the
ordering-
it
probably
tries
both
and
there
must
have
been
some
cases,
whereas
canceling
before
the
completion,
which
is
also
valid,
like
I've
seen
this
happening
around
here,
also
like,
depending
on
which
wins
it
might
be
complete
or
canceled
both
for
success
cases
yeah.
I
don't
like
the
reactive
streams,
but
very
much.
C
C
Oh,
we
I
decided
to
at
least
for
for
halo's
pr,
which
is
the
arbitrary
method
not
like
webflex
stu.
A
C
To
add
an
canceled
attribute.
C
Yeah,
it
looked
like
we
had
some
precedent
there
from
grpc
and
I
didn't
want
to
create
a
the
original
pr
had
creating
a
an
exception,
completing
it
with
an
exception,
and
that
made
me
nervous
because
of
creating
a
building,
a
stack,
creating
the
exception
on
the
potentially
on
the
hot
path.
I
don't
really
know
what
cancelled
I
don't
understand
cancelled,
so
I.
D
D
D
D
C
All
right,
then,
have
a
good
have
a
good
morning
nikita.
Well,
I
would
say:
I'm
jealous
I'm
jealous,
like
you
have
the
whole
day
for
in
front
of
you,
except
that
really
about
my
favorite
time
of
day
is
going
to
sleep.