►
From YouTube: 2021-09-17 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
B
A
Yeah
yeah
he's
got
some
really
good
results.
Finally,
I
mean
really
the
probably
useful
results.
I
guess
would
be
good
and
it's
really
excellent
results
from
the
point
of
view
of
the
open
telemetry
java
agent.
Actually
it's
the
splunk
distro,
but
it's
we
don't
have
any.
I
don't
think
we
have
anything
fancy
in
the
config
that
he
was
running
out
pretty
much
equivalent,
but
overhead
was
extremely
low
for
his
test.
Using
the
spring
tech
clinic
api.
B
I
saw
the
numbers
yeah.
I
definitely
want
to
dive
into
that.
More
do
you
know
how
stable
those
results
like?
Was
that
just
a
one
time,
I
think
he
ran
them
like
five
times,
okay,
and
that
was.
A
A
A
B
A
B
That's
why
I
want
to
dig
in
it
didn't
match
my
my
experience,
but
I'm
looking
forward
to
figuring
out
what
yeah.
What's
the
difference
and
scenarios
and
that's
why
I
mean
that's,
why
nobody
publishes
benchmarks
for
apm
tools,
because
it
so
depends
on
so
many
factors.
A
B
B
A
So
the
only
thing
running
so
I
was
running
in
docker
on
the
host
on
the
host
with
the
agent,
the
application
and
the
agent
was
just
the
k6
driver
and
the
application
and
the
database
and
the
collector
were
taken
off
to
a
separate
host,
which
I
think
is
good
and
also
I
don't
know
whether
people,
whether
people
have
generally
done
that.
But
I
think
that
was
a
very
good
way
to
stabilize
things
a
little
bit.
B
Yeah,
I
wouldn't
explain
my
the
differences
for
me
because
we're
not
using
I'm
not
using
the
collector.
B
Also,
do
you
know
what
hardware
was
that
on
the
github
actions?
Hardware,
no,.
B
Do
you
know
what
the
specs.
B
A
Even
if
everything
is
mostly,
I
o
bound.
The
threads
are
just
fighting
for
the
work
so
much.
C
A
Yeah,
this
was
something
that
me
jason
and
I
spent
most
of
wednesday
afternoon
kind
of
working
through
these
issues,
wednesday
afternoon
tuesday
afternoon
I
don't
remember,
was
just
like
looking
at
results,
because
we
were
actually
the
first
time
he
ran
it
on
two
cores,
the
having
the
agent
there
made.
It
made
all
of
the
made
all
of
the
things
better.
A
We
made
it
better.
I
think,
because
everything
slowed
down
enough
with
the
agent
there
that
it
was
everything
was
able
to
keep
up.
So
all
of
those
all
of
the
stats
were
better
when
the
agent
was
running
and
we're
like
this
makes
no
sense
what's
going
on,
but
we
were
really.
We
were
pegging
the
cpu
we're
pegging
the
cores,
and
it
was
just
it
was
yeah
it
wasn't.
The
data
wasn't
useful.
B
Yeah,
that's
a
good
question
actually
like
that
whole
thing
of
test
measuring
when
you're
measuring
average
response
times
versus
throughput
yeah.
A
A
A
But
I
don't
remember
exactly
what
the
throughput
of
the
throughput
was
well
yeah,
5000
iterations,
but
I
don't
remember
he
did
measure
the
throughput.
You
can
get
it.
I
think
out
of
the
mean
request
time
or
the
iteration
mean
time
or
something
by
dividing
something
by
something
else,
but
anyway
I
should
let
jason
speak
to
it,
because
it's
his
it's
his
thing
and
I
don't
remember.
We
worked
a
bunch
on
it
on
tuesday,
but
I
don't
know
where
it
all
ended
up
yeah,
the
one
thing
that
was
very
well.
A
I
guess
maybe
it
looks
okay
here,
but
it
was
curious
to
me
that
the
the
network
read
the
network.
Read
numbers
made
very
little
sense
to
me
and
because
I
would
think
that
they're
like
the
reach,
the
agent
is
not
going
to
be
doing
any
network.
The
only
network
reading
the
agent
would
have
been
adding,
I
think,
would
be
just
like
the
response
from
the
grpc
call
for
the
export.
A
Yeah
we
the
again
this
is
not,
he
has
a
whole
bunch.
He
has
a
lot
more
data
than
this
and
there
were
some
confusing
looking
results
around
network.
Read
that
I
don't.
We
did
not
really
understand.
Okay,
but
again,
why
would
it
be
like?
Why
would
it
be
anyway?
The
number
network
read
is
weird
and
I
don't
really
understand
what
it
all
goes
into
it,
but.
B
A
A
B
B
I'm
gonna
kick
off
another
run
here,
compare
yeah.
This
is
awesome.
I
I
mean
I
think
it's
a
really
cool
integration
here
into
the
github
action.
B
Yeah
so
anurag,
we
surreal
made
it
today
and
talked
about
the
his
vision
for
cicd
observability,
so
he
was
at
cloudbees
before
so.
Definitely
has
a
lot
of
cicd
background,
so
I
think
he's
kind
of
looking
for
how
to.
B
Get
momentum
in
open,
telemetry,
broader
than
just
you
know,
landing
the
extension
into
the
contrib
repo.
So
we
pointed
him
to,
I
think,
the
sigma,
the
the
spec
meeting
john.
I
think
that
you
called
out
was
the
right.
B
The
right
place
to
go
next.
C
A
B
C
A
B
A
B
B
A
C
B
Well,
yeah,
I
guess
we
could
shade
it.
No,
we
could
shade
it
right
because
we
could
shade
caffeine
usage
of
sun
misc,
unsafe.
B
C
B
C
B
A
A
C
C
B
A
C
Package,
yeah,
that's
defined
by
yeager
notice,
yeah
right
so
like
initially
we
would
have
edited
the
profile,
so
it
so
it
generates
into
our
own
package
and
then
it
would
have
been
fine,
but
we
didn't
so.
We
were
generating
this
jager
package
in
the
public
api,
but
now
we
moved
into
an
optional
artifact
and
so
people
for
some
reason,
are
using
it
and
I
suspect
the
probability
is
very
low,
but
if
they
are,
then
they
can
add
the
dependency
which
so
far,
we've
considered
some
compatible.
I
think
it
is.
A
Yeah,
I
think
that's
I
think
I
think
you're
right
and
like
yeah,
my
guess
is
no
one's
going
to
be
using
the
jaeger
proto.
But
you
never
know
like
people
might
exactly
say
our
publishing
of
the
proto
inside
our
artifact
I'd
rather.
C
A
Did
did
josh
take
a
look
at
it.
I
don't
remember.
A
I
don't
know
if
you
so
you.
If
you
saw
I
put
in
a
pr
to
upgrade
some
dependencies.
It
was
interesting.
We
can't
upgrade
the
junit
bomb
yet,
even
though
there's
a
more
newer
one
published,
because
we
have
other
artifacts
that
depend
on
an
older
version
of
j
unit
bomb
and
they
depend
explicitly
on
well.
Actually
I
don't
know
that
they've
defended
exclusion,
jade
bomb,
but
they
depend
on
the
old
legacy
j
unit,
four
versions
of
things
that
haven't
been
published
that
are
in
the
new
old
one,
not
the
new
one.
A
C
B
C
A
C
A
A
Later,
actually,
it
didn't
show
up
when
I
ran
the
ran
the
gradle
test
to
look
at
dependency
upgrades
available.
A
Yeah
yeah
one
thing
that
I
picked
up
that
I
could
not
figure
out
where
it
even
where
the
dependency
comes
from,
is
the
kotlin
bomb
so
that
it
could
be
upgraded.
But
I
couldn't
figure
out
even
where
we'd
declare
a
version
of
the
kotlin
bomb.
C
A
A
C
A
Oh
hey,
I
noticed.
A
So
that
pr
for
the
the
log
adapter
and
none
of
us
caught
that
that
that
ron
had
basically
put
in
forced
the
dependency
on
otld
common
on
the
log
the
extent
the
alpha
log
extension,
which
we
definitely
cannot
do
so
the
log
adapter
pr
like
five
people,
approved
it,
and
then
I
was
like
wait
a
second.
This
is
not
right,
we're
putting
a
dependency
in
here.
That
definitely
does
not
belong.
A
C
C
C
A
C
A
A
C
Yeah,
probably
I
guess
yeah,
it's
probably
a
bit
early
for
that.
That
makes
sense
like
we
would
probably
do
it
later,
but
not
right.
Now.
I
think
that's
fair,
like
once.
We
have
multiple
exporters
for
logging,
then
we'd
probably
need
to
move
it
to
some
shared
location
and
that
might
be
utility
common
or
we
might
have
had
a
different
scheme
set
up
by
then
we
don't
have
to
do
that
right
now.
That's
true.
A
C
Yeah
I
mean
the
fact
is
to
change
this,
to
compile
only
dependency
for
one
line
change
and
then
we
can
live
with
it
for
a
while
anyways,
because
then
there's
no
downstream
effect,
because
it
would
still
like
the
exporter
would
still
depend
on
the
sdk,
probably
so
like.
Even
maybe
all
four
of
these
could
be
compiling.
I
don't
know
how
our
things
are
set
up,
but
otf
common
doesn't
necessarily
have
to.
C
A
B
Oh,
I
wanted
to
ask:
what
did
how
what
do
you
run
to
look
for
dependencies.
C
C
A
B
C
B
Oh
this
one,
oh,
I
saw
that
and
I
immediately
just
thought:
caffeine,
all
right
actually.
A
Is
has
dependable,
stopped
running
or
something
altogether,
because
I
haven't
seen
a
depend
about
vr
for
months
now,.
C
C
B
A
C
C
Yeah,
I
think,
that's
better
because
it
sends
a
different
pr
for
each
dependency.
It's
this
and
it
might,
I
think,
even
if
you're
the
configured
for
every
month
like
if
I
release
it's
a
bit
delayed
or
it
gets
early
or
something
it
might
not
line
up
every
week
is
just
too
annoying
because
there's
no
point
updated.
Definitely
we
don't
release
anyways.
C
C
A
C
B
C
B
C
A
No,
no,
I
think
this
was
just
a
general
comment
that
that
more
and
more
more
and
more
people
will
need
to
release
mr
jars
because
java,
you
know,
java
8
is
now
so
old,
and
yet
everyone
tons
of
people
are
still
using
it.
So
I
think
it
was
spring.
Sex
was
just
a
it's
a
canary
in
the
coal
mine
about
having
to
do
this.
It
sounded
like
from
what
ken
said.
C
B
Yeah
we
can
like
we,
we
have
java
11
instrumentation,
we
just
the
instrumentation
module
we
target
to
java
17
java,
11
or
java
17,
and
then,
when
we
load
it
at
runtime,
it
throws
a
like
a
class
format
exception
that
we
just
silently
ignore.
B
Actually,
we
don't
it's
not
too
silent,
it's.
It
says
something
about.
We
actually
log
it
at
debug
level,
and
so
I
have
actually
had
two
customers
that
asked
like
they
thought
that
we
had
compiled
our
whole
agent
against
11,
because
it
would
print
out
that
it
couldn't
run
the
http.
C
Annoyed
by
the
random
debug
blog
messages,
like
the
neti
thing,
is
also
like
every
single
customer
and
they
have
a
problem
with
agent
like
I'm
getting
this
exception
and
they
copy
paste
that
medi
or
grpc
auto
detect
whatever
like
that
happens
every
time.
So
maybe
it's
finally
time
to
do
something
for
our
debug
vlog.
So
it's
easier
to
provide
tech
support
because
yeah.
B
It's
just
yeah.
I
just
turned
off
just
in
our
last
release.
Here
I
just
we
hit
the
debug
log
level
that
you
set
for
our
agent.
I
don't
set
open
telemetry
to
debug,
so
it
only
applies
debug
to
our
code.
So
not.
B
C
C
B
C
Or
I
mean,
of
course
another,
but
then
yeah
this,
I
don't
know
if
we
can
actually
do,
but
if
they
weren't
full
stack
traces
like,
for
example,
I
couldn't
find
the
grpc
implementation
river
so
that
always
like
the
stack
traces.
That's
why
it
looks
so
much
like
a
failure
to
a
customer.
It
was
just
a
message.
I'm
sure
that
would
have
been
okay,
though
so
maybe
a
custom
login
handler
to
do
something.
That
is
another
option.
C
B
A
C
B
A
Well,
this
is,
is
it
proposed
both
by
oracle
and
adopt,
I
think,
to
go
to
a
two-year
cycle
rather
than
a
three-year
cycle.
A
B
B
An
agent
that
a
delaying
initialization
of
something
in
an
agent
by
instrumenting
public
static,
void
main
and
calling
like
an
internal
agent
init
method
where
we
could
do
things
so
this,
for
example,
just
you
know,
does
the
only
once
and
then
we
could
do
init
in
here.
Runtime.
C
B
Right
before
we
run
the
customer's
main
and
that
works
fine
as
mischief
oh
great,
is.
C
B
B
Yeah,
so
it's
needed
because
we
need
somewhere
to
start
up
all
our
threads
like
the
and
exporters.
C
B
Oh
well,
pre-main
pre-main
doesn't
get
executed.
So
it's
interesting
pre-main
gets
executed
when,
while
you're
building
the
image
for.
B
C
C
B
Maybe
right
yeah
yeah,
so
the
main
their
main
it's
going
to
it'll
call
during
compilation,
it
calls
maine
and
com.
Does
it
call?
Yes,
because
that's
how
I'm
instrumenting,
because
it's
it's
actually
we're?
Actually,
I
was
instrumenting
during
the
the
comp,
the
native
compilation
phase,
so
that
if
our
byte
code.
C
B
C
Because
I
remember
like
I
think
it
was
ken
who
was
mentioning
needing
some
sort
of
lazy,
initialization,
probably
outside
of
the
context
of
agents,
so
that
I'm
trying
to
sort
those
two
topics
together,
I'm
just
a
little
bit
confused.
I
think,
but
if
I
remember
that
coming
up-
and
so
I
assumed
that
there
must
be
some
mechanism
already-
maybe
that's
a
quickest
mechanism-
maybe
another.
A
C
B
And
you
can,
do
you
opt
in
or
opt
out?
I
think
maybe
you
I
think
by
default.
It
doesn't
run
any
of
them,
but
you
can.
C
B
In
and
of
course,.
C
B
A
The
things
that
ben
tried
to
hammer
home
is
we
shouldn't,
try
to
work
around
the
growl,
any
graw
limitations.
We
should
log
bugs
against
for
all
that
they
need
to
do.
They
need
to
do
things
better.
B
Yeah
not
to
be
scared
off,
but
because
I
was
saying
that
I
started
reading
all
the
things
that
you
couldn't
do
in
grawl
about
class
loaders
and
I
basically
stopped
working
on
the
project
because
it
freaked
me
out
because
I
like
we,
you
know
we
inject
classes
into
class
loaders
at
runtime,
like
the
more
I
was
reading
about
they're
more
like
this
is
there's
no
class
loaders.
How
do
we?
Even
so?
Yes,
so
he
was
like
no?
No!
No
don't
we
need.
We
need
this
we'll.
B
B
He
was
at
a
new
relic
before
and
before
that
at
the
j
clarity,
which
did
also
performance
monitoring,
which
is
where
microsoft
acquired
a
couple
of
good
java.
We,
where
we
acquired
basically
what's
now
our
java
internal
java
engineering
team.
B
Oh
yeah
jason
asked,
I
think,
jitter
meaning
like
the
different.
B
Random
jitter,
oh,
is
that
circuit.
Thank
you.
A
Anyway,
ben
was
saying:
there's
actually
two
pieces
of
the
of
hot
spot,
the
hot
spot
jitter,
which
is
what
people
generally
call
it
actually
there's
the
c1,
which
is
all
the
stuff,
that's
pro
that
it
can
prove
and
then
there's
a
c2
which
is
kind
of
the
stuff
it
tries
to
guess
at
and
then
has
ways
to
back
out
when
it
is
like
monomorphic
dispatch,
yeah.
A
B
So
yeah
and
that's
I
had
heard
that
before
that
the
while
the
startup
was
obviously
way
faster,
that
your
overall
performance
was
actually
not
as
good.
A
B
A
Stuff
like
that,
then
you
probably
will
gain
more
from
not
running
running
time,
but
if
you
have
a
fairly
straight
ahead
app
that
builds
everything
up
front
and
just
does
its
work
and
doesn't
do
anything
kind
of
very
dynamic.
Then
you'll
probably
do
very
well
with
I
mean
it's
like.
If
you're
going
to
compete
with,
go
like
that's,
the
most
important
part
right
go
doesn't
have
any
sort
of
yeah.
That's
good.
C
B
Yeah
I
started,
I
learned
just
a
tiny
bit
of
assembly,
whereas
oh,
I
think
I
actually
have
it
on.
B
I
think
it's
actually
a
pinned
now,
of
course,
I
lost
it.
B
Let's
find
some
fun.
Yes,
it
took
me
forever
to
like
watch
some
videos
to
try
to
understand
what
on
earth
the
stuff
meant,
but
they
do
actually
it
prints
out
a
pretty
nice.
I
mean
this
stuff
is.
C
A
I
was
actually
kicking
beehives
in
the
log
sink
this
week
because
they
were
yeah
anyway.
B
B
I
think
the
button
should
work
and
I
think
I
understand
I've
been
I've
been
messing
with
the
build
scripts
lately.
So
I
actually
think
I
understand
things
better.
You.
B
B
Yeah,
it
was
it's
that
gradle
plug-ins
is
backed
by
this
is
still
is
this
store?
Yes,
our
problem.
Hey.
A
A
B
B
Smoke
tests,
let's
find
the
ones
that
we're
running
on.
B
I
should
see
if
adopt
has
published
17
yet
because
or
no
eclipse,
because
they
had
they
don't
publish
pre-releases.
Unfortunately,
maybe
they
will
in
the
future,
but
the
the
docker
community.
One
is
they're,
really
good
at
publishing
all
the
early
access
releases.
B
So
I
was
able
to
miss
that
this
was
running
against
the
early
access,
yeah,
and
so
now
I
think
they
did
they
did.
I
looked.
They
did
publish
17
already,
so
this
should
be
running
against
the
real
release
already,
so
is
it
out
17
it's
up
today,
17
was
out
on
tuesday
I
lost
my
video,
so
you
can't
see
my
wild
hand
waving.
B
And
eclipse
yeah
the
this
is
unfortunate,
the
there's
no
17
here
yet
just
because
this
was
also
what
like
the
github
actions
right
is
based
on
adopt
or
oracle.
I
think
where
are
two
options.
B
I
don't
see,
is
this,
do
you
know
which
can
you.
B
B
So
yeah
so
at
least
we
have
that's
kind
of
why
I
had
just
put
in
this
pr
a
few
days
back
just
so
that
we
could
say.