►
From YouTube: 2021-03-19 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
C
B
Yeah
you're
doing
I
saw
there
was
some
tension
in
the
community
repo.
C
If
you
haven't
pointed
out
correctly,
I
would
have
just
never
noticed
we
were
using
the
aws
yeah
that
just
didn't
come
to
mind
and
like
that's
how
my
sample
apps
generally
were
written,
so
I
just
copied
them
over.
It's
like
oh
yeah.
We
shouldn't
be
using
destroys
here,
but
what
do
we
do
then?
Right.
B
C
B
It
seems
like
kind
of
like
the
collector.
C
C
C
C
C
B
C
A
B
Yeah
I'm
trying
the
new
trick
of
just
sharing
a
small
window
on
a
big
screen.
B
Oh
I
passed
on
your
john,
your
your
github
workflow
tips
to
my
manager.
Today
he
was
really
excited
because
he
was
drowning
in
github
notifications.
B
But
somehow
none
of
us
knew
about
her
trip,
but
I
believed
in
like
I
always
saw
it
there,
but
I
was
like
this
is
not
gonna
work
very
well.
C
C
A
C
C
B
B
B
So
many
fun
little
gotchas
and
problems
yep.
B
Yep
yep,
oh
good,
stuff
yeah.
I
have.
I
have
one
internal
support
that
came
in
today:
we're
crashing
some
app
because
we're
adding
the
that
interface,
the
field
back
provider
interface
twice
to
some
class,
I
think,
related
to
some
proxying.
B
I
remember
seeing
that
the
datadog
folks
were
doing
some
work
in
their
repo
a
couple
months
back
related
to
that.
So
I'm
going
to
check
that
out.
B
Potentially,
pull
that
merge
that,
if
it
resolved,
if
it,
if
it's
the
same
thing,
oh
yeah,
this
was
interesting.
Just
the
with
java
16,
the
I
guess,
the
all
those
reflections
warnings
that
we've
been
ignoring
are
now
getting.
C
B
C
C
B
C
C
Really
I
thought
groovy
does
this
all
the
time,
though
I
guess
we
don't
reach
into
the
jdk.
So
maybe
that's
why,
okay?
That
must
be
why
we
do
a
lot
of
stuff,
but
probably
not
jdk
really
did
so.
B
The
setup
I
my
I
I
was
asleep
at
the
beginning
of
the
meeting
and
luckily
somehow
woke
up
without
my
alarm
at
just
after
the
meeting
started,
but
john
jason
am
I
stating
this
correctly?
Is
this
the
same
thing?
I
was
assuming
it's
that
reflect
the
reflection
error
warnings
that.
E
I
didn't,
I
think
it
don't
think
it
was
reflection.
I
think
it
was
using
stuff
like
unsafe,
which
they've
said
nobody.
Can
you
like
all
those
internal
classes
that
you're
not
supposed?
You
were
never
supposed
to
use,
but
people
used
all
the
time
like
unsafe.
They
basically
made
it.
So
you
cannot
use
that
stuff
anymore
without
some
extra
switches.
E
C
E
So
the
the
real
difficulty
that
we
kind
of
came
down
on
with
this
issue
is
how
so
we
can
disable
testing
on
java
16
for
some
set
of
libraries.
E
C
B
C
C
C
D
B
C
Yeah,
I'm
sure
that's
a
few
lines
of
extra
if
statements
in
our
groovy
files,
but
so
that
this
line
from
nikita
does
it
like
he's
saying
how
do
we
run
later
step
test
against
java
16.
B
D
B
B
Is
this
the
the
oracle
distribution.
C
B
C
B
D
D
Adopt
is
going
to
eol
lts
java
8,
it
says
at
least
may
2026,
so
they've
got
a
lower
bound.
B
B
C
B
C
B
Like
working
out,
everything.
D
B
E
I
know
there
there's.
I
think
this
is
at
least
fairly
public
knowledge
that
there's
some
there's
a
universe,
some
university
folks
in
estonia
who
are
going
to
be
working
on
master's
projects.
I
think
to
do
compile
time
to
get
compile
time.
Instrumentation
working.
E
I
don't
know
where
that
is
or
what's
happening.
I
haven't
really
been
following
along,
but
I
know
somebody
from
one
of
our
estonian
colleagues
has
been
leading
that
up.
D
B
Yeah,
it's
cool
because
that'll
hit
two
two
areas
at
once,
potentially
for
us.
B
Yeah,
I'm
surprised
how
much
I
see
about
like
growl
and
corkus.
It
seems
like
a
lot
of
people
are,
or
at
least
the
people
who
like
to
post
stuff,
are
trying
that
out
a
lot.
E
So
you
can
compile
kotlin
down
to
native
native.
I
don't
know
how
it
works,
but
you
can.
I
think
the
idea
is,
if
you
want
to
you,
want
a
high
performance,
you
can
compile
your
kotlin
down
to
native
code.
B
So
this
is
not
dalvik.
This
is
what's
right,.
A
D
E
B
I'm
glad
that
something
came
along
to
jvm
alternative
language
that
is
more
popular
than
scala.
E
A
C
C
C
C
To
me,
yeah
yeah,
I
think
you
have
to
only
use
references
if
you
have
any
pointers
in
there,
maybe
not,
but.
B
C
E
A
E
B
Ted
is
trying
to
sort
of
organize
like
talk,
drive
the
discussion
of
how
to
deal
with
exploding
numbers
of
instrumentations
sort
of
I'm
thinking
long
term
like
even
if
we,
for
example,
for
us,
you
know
we
at
some
point.
We
could
split
off
and
have
a
contrib
repo,
but
even
like
long
term.
B
There's
there's
some
limitations
there
unless,
as
I
think
it
was
materials
brought
up
the
or
nikita
that
if
we
can
push
more
native
library
instrumentations
that
potentially
keep
our
curve
from
going
up
up
and
forever
up
yeah.
D
D
But,
like
you
know,
I
feel
like
there's
a
big
core
active
actively
developed
chunk,
especially
open
source
junk.
That
probably
is
a
prime
candidate.
C
So
with
my
experiences
with
working
with
like
this
couch
base
and
vertex
on
their
instrumentations,
when
I
found
them
instrumentation
is
such
a
hard
thing.
I
think
we
all
know
that
I
so
I
want
to
be
sort
of
optimistic
if
we
can
present
a
good,
instrumentation
api.
It's
still.
Okay,
I'm
not
sure,
like
the
concepts
of
instrumentation
is
still
so
hard
context.
Propagation.
What
not
is
every
library
really
equipped
with
the
skills
to
do
this,
and
maybe
it's
just
like?
C
Maybe
the
solution
is
then
like
they
still
own
it,
but
they
do
get
one
some
of
us
in
at
least
as
consultants
or
code
reviews,
or
something
to
make
sure
at
least
they
have
some
help
getting
the
initial
stuff
going,
because
most
libraries
will
have
trouble
with
this.
I
think
and
the
result
will
be
bad
instrumentation
and
that
could
be
bad
for
users
also
compared
to
us
having
written
in
the
first
place.
C
B
A
little
bit
of
discussion
about
like
so
if
third
parties
wanted
to
write
java
agent
instrumentation
how
we
would
pull
those
in,
and
so
one
option
is
via
the
hotel
java
agent
extensions.
Basically,
we
just
need
to
get
that
onto
the
agent
class
loader
into
the
agent
class
loader,
but
also
kind
of
the
question
was
asked:
can
we
can
we
include
that
instrumentation?
Like?
Could
the
user
add
that
to
their
pom
file?
B
Add
that
into
their
app
and
then
we
could
dynamically
load.
That
java
agent
instrumentation
from
the
class
path,
which
is
trickier
and
kind
of
depends
on
trickiness.
D
C
B
Right
right,
yep,
I
like
this,
and
this
can
be
a
directory
or
a
specific
jar,
and
you
could
just
drop
in
multiple
instrumented
third-party
java
agent
instrumentations.
If
you
wanted,
we
can
pull
those
in.
B
A
B
Sdk
stuff
anything:
what's
the
finally
we're
ready
to
merge
the
deck
stand.
Processor.
C
E
It
is
nice
to
be
able.
I
mean
jmh
now
has
async
profiler
support
built
into
it,
but
it
I
can't
get
it
to
generate
flame
grass.
It
craps
out
on
generating
flame
grass,
but
I
can
generate
like
html
tree
views
of
the
stacks
and
things
anyway
just
trying
to
get
a
good
workflow
for
doing
profiling
work,
and
I
can't
say
I've
got
anything
amazing
yet,
but
still
working
on
it.
C
E
Yeah
there's
just
a
lot
of
moving
parts
and
I
would
love
to
be
able
to
run
all
this
stuff
in
the
ide
like
an
idea
and
I've
got,
I
can
do
everything
except,
I
can't
run
jmh
benchmarks
and
profile
them
at
the
same
time
with
an
idea.
I
can
only
do
that
from
gradle,
so
it's
not
the
end
of
the
world,
but
sometimes
it's
nice
to
be
able
to
just
stay
stay
in
the
ide
a
little
bit
more
anyway.
E
B
What
I
remember
doing
was
with
jmh
adding
the
jfr,
just
as
the
jvm
args
to
the
jmh,
the
fork
that
it
passes
along
and
then
running
like
a
fairly
long,
continuous
profile.
I
mean
a
continuous
run,
the
benchmark
for
like
30
seconds
or
something
long
enough
that
and
then
just
looking
at
the
jfr
stuff.
E
As
interesting,
I
was
showing
jason
yesterday
with
async
profiler.
I
think
I
caught
a
jit
hat
like
jit
inlining
happening
in
the
pro
like.
I
could
see
it
in
the
profiler,
because
I
can
see
stacks
from
two
different
like
the
same
code
path,
but
two
different
stacks,
one
of
which
had
looked
what
looked
like
inlined,
an
inline
method
in
it,
which
I
thought
was
kind
of
cool
to
actually
be
able
to
observe
the
changes
the
jit
had
made
in
the
profiles.
B
Yeah
with
that
async
stuff,
our
getting
really
like
get
because
it's
not
precise,
like
you
get,
you
can
get
drift
god.
What
was
it
it
really
confused
that.
B
B
Well,
I
was
using
jfr
with
the
they
have
some
flags.
You
can
turn
on
to
get
async.
E
Oh,
so
it
would
track
stacks
across
async
boundaries.
Is
that
what
you
mean?
No,
so
are
you
talking
async?
Do
you
mean
the
like
async
profiler,
the
actual
async
profiler
project.
B
Okay,
which
is
what
not
you
not
profiling
at
save
points.
B
B
Yeah
so
jfr
has
options
you
can
enable
essentially
that
same
async
behavior
where
it
doesn't.
You
have
that
save
point.
B
B
B
I
think
it's
called
the
async,
because
I
think
the
it's
like
the
underlying
jvm
call
is
called
like
get
async
or
something
to
get
an
async
stack
trace
from
the
thread.
E
So
there's
some
way
that
you
that
call
can
give
you
like
a
broken.
A
broken
stack.
That's
not
actually
a
real
one,
yeah
interesting.
It
doesn't.
That's
not
surprising,
like
I
don't
think,
there's
any
like.
I
bet
there's
some
weird
glitches
that
could
happen
in
there,
but
I
think
this
this
look.
These
looked
like
legit
two
legit
stacks,
but
one
of
them
had
it
it
had.
Basically
you
could,
you
could
tell
it
was
in
line
because
there
was
a
inside
the
jdk.
E
There
was
a
switch
from
calling
something
that
was
like
generated,
something
something
yeah
it
was.
It
went
to
something
called
generated
from
something
that
was
called
not
generated
anyway,
so
it
looked
like
and
then
you
could
even
see
that
in
the
stack
there
was
a
method
that
had
disappeared.
So
it
looked
so
the
stack
looked
exactly
the
same
except
the
top
was
different,
had
been
renamed
and
then
a
method
on
inside
it
actually
disappeared.
B
E
Well,
it
showed
up
as
two
different
stacks,
so
you
kind
of
had
to
manually
add
the
numbers
together.
If
you
get
like
what
the
total
the
total
number
of
pro
total
number
of
stacks
that
were
for
that
actual
code
path,
so
anyway,
it
was.
It
was
just
kind
of
interesting,
and
I
think
that
I
mean
I
can't
I
don't
have
proof,
but
it
certainly
looks
like
that's
what
was
going
on.
B
Yeah,
if
you're
into
looking
like
the
inlining
stuff,
this
tool
is
actually
pretty
easy
to.
Like
surprisingly,
like
you,
it's
like
gooey,
you
run.
B
You
enable
the
legit
like
diagnostic,
like
vlogging,
which
is
huge
and
painful
to
comb
through,
but
then
this
gui
tool
you
can
open
up
and
click
into
your
class
and
see
like
the
stuff
that
it's
done
in
lightning
and
I
haven't
solved
any
problems
with
it.
But
it's
cool
to
look
at.
E
D
B
E
I
think
the
thing
that
I
took
away
from
that
was,
even
if
you,
even
if
you
let
your
benchmark
warm
up
for
30
seconds,
you
can
still
get
jits
happening
during
the
actual
benchmark.
Benchmark
runs
like
30,
like
it's
there's,
no
real
way
to
tell
how
long
you
need
to
let
things
warm
up
to
get
rid
of
jit
effects.
E
It
was
in
this
benchmark
that
was
contributed.
That
is,
it's
not
maximize
the
cpu.
It's
a
steady
throughput
benchmark.
B
E
E
So,
for
example,
in
our
old
in
the
existing
batch
sorry
batch
band
processor
benchmark,
our
goal
was
more
to
just
see
how
much
we
could
push
through
like
what
throughput
could
we
get
to
with
the
bass
band
processor,
which
is
a
different
question
than
under
normal
load?
How
much
does
cpu
does
this
dashband
processor
itself
use.
B
E
Not
necessarily
because
it
especially
with
the
bachelor
processor
things
get
super
tricky,
because
if
you
try
to
send
as
many
things
as
you
can
into
it,
it'll
start
dropping
spams
and
so
you're,
not
when
you
start
dropping
spans
everything.
Everything
gets
crazy
right.
C
E
B
With
the
jmh
with
for
fleshing
I
mean,
would
you
normally,
you
would
move
the
flush
to
after
the
benchmark
loop?
I
mean
the
timed
component
like
to
have
like
a
setup
teardown
or
something
that
oh.
E
We
probably
wouldn't
bother
to
flush
at
all,
like
there's,
no
reason
to
flush
if
you're,
what
you're
trying
to
mean.
Normally,
you
wouldn't
bother
to
do
a
flush
at
all,
because,
if
you're
having
to
do
a
flush
you're
in
a
weird
situation,
like
maybe
you're
in
a
place
where
you're
having
to
like
you're
in
a
lambda
or
whatever,
and
you
need
to
make
sure
you
shut
it
down
like
get
everything
out,
but
that's
not
a
normal
situation.
E
Okay,
like
it's,
not
a
constant
throughput
situation,
you
would
never
have
to
do
that
flush,
because
everything,
because,
like
the
lambda,
would
still
be
alive,
you
wouldn't
need
to
shut
anything
down
and
if
you're
running
a
server
like
you
wouldn't
bother
to
do
flushes
ever
right.
It's
just
maybe
maybe
when
your
jvm
exits,
but
that's
not
really
part
of
the
benchmark,
and
it
doesn't
really
matter
for
benchmarking.
E
E
B
D
C
E
E
And
you,
you
lose
the
kind
of
built-in
intelligent,
intelligent,
warm-up,
warm-up
cycles
and
yeah,
like
I
actually
tried
to
do
this
yesterday.
Also
exactly
that
was
like
what
if
I
didn't
use
jmh
and
just
wrote
my
own,
I'm
like
I
have
to
like
reproduce
jmh,
basically
to
do
that
and
then
that's
that's
a
big
name.
E
Cool
I've.
I
also
found
with
my
profiling
today
that
these
stacks
that
spring
generates,
which
I'm
sure
we
all
know
but
they're,
crazy,
they're,
like
just
so
gigantic.
Have
you.
B
Have
you
seen
react
reactive,
rxj,
yeah.
Those
are
even.
C
E
C
D
All
that
tells
us
is
that
that's
the
way
to
find
it
out
seven
years
ago.
B
There's
a
cool
site
like
why
is
it
not
that.
D
B
Seo,
I
think
chris,
who
codes.
Actually,
I
think
that
same
person
chris,
who
codes
jvm.
A
B
A
B
Recently,
so
what
did
we
want
to
find?
We
wanted
xss
stack
size.
D
B
B
B
That's
that
would
be
some
fun
stack.
Overflow
problems,
errors.
A
C
B
So
I
didn't
quite
follow
what
exactly
he's
saying.
Oh.
C
B
But
do
you
understand
what
is
wrong
about
the
pictures.
C
C
It's
clock
skill,
I
don't
I
mean
with
the
timer
that
we
propagate
down
through
the
appearance
bands,
it's
the
same
clock
or
whatever
I
wouldn't
expect
the
timings
to
be
wrong.
So
it's
more
likely
to
start
spammed
and
I
I
didn't
look
at
the
sample
project.
So
probably
the
first
thing
to
see
what
instrumentation
they're
using
then.
E
D
Yeah
I
mean
so
we
I
mean
so
a
go
client
right
in
that
small
example.
The
go
client
is
probably
finishing
its
span
when
it
gets
the
final
bite
right
like
it's
doing
an
http
request
when
it
sees
the
final
bite
of
the
response.
It's
like
got
my
response
and
it's
possible
that
there's
still
stuff
happening
on
the
java
side,
like
we've
sent
the
final
byte
out,
but
we
haven't
finished
whatever
call
stack
and
cleanup
is
required
for
the
request.
D
D
D
E
D
Or
until
it's
got
its
first
bite
on
the
wire
right
or
until
it's
like
completed
sending
headers-
or
I
don't
know
like
I
said,
I'm
speculating,
but
that
one's
definitely
weirder
than
the
ending
case.
E
All
right,
well,
I'm
going
to
take
off
one
one
last
fun
thing
that
I
discovered
today
that
that
is
that
the
accuracy
of
the
scheduled
thread
pool
executor
is
not
very
good,
like
five
milliseconds,
five
milliseconds,
plus
or
minus
on
your
your
delay
very
common
and
easy
to
see,
I
was
actually.
The
thing
I
was
surprised
is
that
it
often
executed
before
the
like.
If
you
say
I
want
a
fixed,
fixed
rate
of
200
milliseconds,
it
would
often
execute
at
195,
milliseconds
or
the
time
it
actually
like
before
your
request.
E
C
E
Yeah,
it's
really
sloppy.
I
was
really
surprised
at
how
like
I
was
really
surprised
that
it
would
be
faster
than
your
time,
like
I
thought
sure.
If
it's
a
little
delayed
like
maybe
you
know,
the
kernel
didn't
wake
up
the
thread
until
the
right
time
or
whatever,
but
yeah
it
was.
It
was
often
before
it
and
jason,
and
I,
like
jason,
wrote
a
little
thing
to
test
this
and
it
was
absolutely
true
so.
C
E
C
E
E
All
righty
well
take
it
easy,
have
a
good
night.
I
will
not
see
you
all
for
probably
for
a
week
because
I'm
on
vacation
next
week
so.
A
E
E
Yeah,
I
think,
like
the
first
week
of
april,
it's
a
good
time
for
one
that
one
all
right
cool
have
a
good
one.
See
you
all
later
bye.
D
C
C
Yeah,
I'm
gonna,
I'm
sure
this
is
still
gonna
take
a
while
to
get
into
something
merchable.
So
yeah,
it's
good.
I
think
there's
some
things
about
it
that
I'm
still
not
super
satisfied
with,
but
it
yeah
some
one
thing
I
didn't
like
was
these
insanely
big
constructors,
but
then,
when
I
think
about
it,
is
it
that
different
having
a
big
constructor
versus
having
a
bunch
of
protective
methods?
I
guess
they're
the
same
thing,
so
maybe
it
doesn't
matter.
C
C
C
B
B
Oh
sure,
for
other,
for,
like
a
vertex
or
couch
box,.
C
I
think
it
gets
them
the
customization
they
need
and
as
opposed
to
our
current
approach,
where
there's
no
customization
at
all,
so
that
probably
worked
out
well,
I'm
still
want
to
think
a
bit
more
on
if
the
instrumentation
experience
can
get
a
bit
better,
it's
not
bad.
Maybe
it's
it's
only
this
constructor
that
really
stood
out
as
being
really
annoying,
because
I
had
to
write
it
because
that
then
propagates
into
the
builder,
like
I
had
this
huge
constructor
invocation
and
started
getting
all
these
arguments
together,
and
so
it
was
a
bit
annoying.
B
C
C
B
B
Yeah
yeah
we'll
keep
our
fingers
crossed
for
to
hear
back
from
nikita
about
muzzle.