►
From YouTube: 2021-08-13 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
B
B
B
I
did
see
it
well,
I
saw
the
title
of
that.
I
have
not
read
it
yet.
I
think
I
saw
john
was
john
commented
on
it,
so
I
knew
it
was
in
good
hands.
Yeah.
A
B
C
B
B
B
A
Okay,
yeah
mainly
because,
like
with
the
finalizer,
I
don't
know
if
it's
you
could
use
a
cleaner
which
is
effectively
weak
map,
but
that's
java
9.
I
guess
on
java
8
the
only
way
to
have
data
when
you
get
notified
for
garbage
functions
to
use
the
map.
Otherwise
you
just
have
a
null
object
with
the
reference
queue.
A
A
C
A
A
This
one,
that's
the
context
debugging
for
foreign
does
is
just
flushes
the
span
using
the
spam
context
to
the
back
end,
which
is
probably
you
know
like
we
wouldn't
need
to
allocate
the
robos.
I
think
for
this
thing.
Oh.
C
A
C
A
A
B
D
C
A
C
A
C
A
With
the
I
guess,
with
an
attribute
to
deflate,
to
determine
whether
a
spam
that's
experienced
in
process
or
not,
I
guess
maybe
just
not
having
an
end
time.
That
might
be
good
enough.
So
maybe
it
is
possible,
like
I'm,
less
worried
about
sdk
more
about
otlp
itself,
whether
that
models
it
properly.
Oh.
A
A
B
D
B
X-Ray,
when
do
you
export,
do
like
every
so
many
seconds,
export
in
progress
spans
that
are
taking
a
long
time.
B
Because
I
I
did
that
in
glow
root.
C
B
A
B
I
would
glow
root
every
by
default
once
a
minute
or
once
a
trace
exceeds
a
minute.
Then
it
will.
D
B
Or
not
spit
well,
in
glow
root,
this
band,
it's
traces
like
everything,
it
keeps
the
whole
trace
it's
even
worse.
It
keeps
the
whole
trace
in
memory.
B
A
Yeah,
we
I
mean
stick
actually
supports
both
modes,
it
can
flush
bands
every
time
it's
ended
or
it
can
collect
the
trace
now,
and
I
just
get
rid
of
the
feature
that
collects
the
trace,
like
even
I
think
the
implementation
could
be
better.
But
for
now
it's
like
some
reference
count,
type
of
thing
where
once
there
aren't
any
references
to
the
root
span
anymore,
it
gets
exported
or
something
like
that,
and
we
have
thread
safety
issues
and
it
doesn't
actually
work
properly.
D
B
Because
they're
it
was
a
their
back
end
needed
the
whole.
B
B
A
Maybe
I
should
have
added
my
own
context,
which
I
didn't,
but
my
thought
is
especially
to
get
the
sampler
out.
I
wouldn't
think
too
much
in
this
release.
A
B
A
B
B
A
One,
and
so
that's
again
like
this-
is
just
a
thing
to
go
through
all
the
trade-offs.
I
think
the
jscontrip
and
the
go
report
moved
in
the
direction
where
each
artifact
would
be
able
to
publish
its
name,
hyphen
version
as
a
git
tag,
and
so
you
have
okay
yeah
a
lot
of
these
tags,
which
is
a
bit
confusing
like
I
still
prefer
multi-repo.
If
we're
going
to
go
with
that
approach,
but
multi-repo
and
hotel
it's
a
bit
annoying
because
you
always
have
that
stuff.
You
always
have
to
ask
for
repo.
A
A
Yeah
and
with
with
the
fyi
that
yeah,
so
the
sampler
and
the
jms
might
also
be
released
like
even
though
I
don't
care
about
the
gms
matrix,
I
don't
mind
it
being
released.
It'll,
be
I
mean,
assuming
I
assume
the
current
one
is
broken,
because
it's
1.0
by
now
the
produce
change
so
much
that
it
shouldn't
work.
I
think
maybe
it
somehow
does
but
it'll
probably
be
good
to
get
the
released
anyways,
but
I
wouldn't
announce
it
or
anything
like
that.
C
B
D
B
C
D
A
D
B
Into
talking
about
instrumentation
going
over
there
and
the
dependency,
oh,
I
know
because
we
were
talking
about
the
build
orders
and
dependencies
between
repos
and
if
nikita
was
thinking
that
the
java
agent
we
publish
in
the
the
instrumentation
repo,
would
pull
in
contrib
instrumentations.
B
So
I
was
saying
that,
like
kind
of
what
I
was
imagining
was
that
the
contrib
repo
would
have
its
own
java
agent
distro.
That
has
all
of
those
instrumentations
kind
of
like
talked
about
that
before
yeah,
and
so
there
were
thoughts
about
splitting
doing
that
split
sooner
than
later.
B
B
So
the
proposal
to
only
include
to
keep
to
basically
do
away
with
the
all
jar
or
have
the
basic
jar
have
only
the
otlp
exporter
other
experts
in
java
agent
contrib.
That
then
we
kind
of
agreed
that
jaeger
and
zipkin
are.
A
B
So
I
think
the
kind
of
agreement
was
that
we
should
at
some
point
flip
the
meaning
of
the
dash
all
basically
have
the
base.
One
have
the
exporters
in
it
and
have
a
separate
one.
That's
like
dash
no
exporters
or
something
yeah,
but
to
maybe
not
do
that
right
away.
Maybe
do
that
at
the
same
time
that
we
split
out
the
contrib
instrumentation
so
that
there's
not
sort
of
two
breaking.
B
Nikita
and
john
really,
like
the
start,
open
telemetry.
I
o
idea
of
pick
your
own.
C
A
A
C
Well,
ben
so
ben
evans,
from
his
experience,
yeah
from
his
experience
on
other
projects
that,
if
you
the
longer
you
wait,
the
messier
and
harder
it
is
to
do.
A
B
A
A
A
That's
definitely
a
requirement.
I
think
we
don't
want
to
have
to
do
with
that.
Of
course,
people's
on
the
flip
side
once
it's
stable,
then
that's
that's
a
reason
for
it
to
be
stable,
set
people
outside
already
the
right
instrumentation.
So
that's
a
good
point,
but
even
assuming
it
was
stable,
that's
like
do
we
really
get
a
huge
amount
by
putting
it
out
and
I'm
sure
repo
is
still
going
to
take
forever
to
build,
even
with
just
the
tier,
1,
instrumentations
and
so
motivation
is
low,
even
though
it
sounds
like
a
good
idea.
A
A
B
B
C
Yeah,
it's
really
sweet
and
I
can
go
ahead
and
release
it
tomorrow.
I
don't
know
if
there's
anything,
that's
currently
sitting
needing
to
be
merged
beforehand.
It
looked
like
everything
that
was
critical.
That
got
done.
C
Okay,
I'll
I'll
go
ahead,
and
do
it
really
and
then
somebody
will
have
to
bite
the
bullet
of
updating
the
agent
metrics
bridge
for
the
new
api.
A
A
C
A
B
This
is
really
cool,
so
the
quercus
folks
are
working
on
supporting
agent
job
agents
by
basically
freezing
the
at
build
time.
Freezing
the
jvm
after
the
pre-main
runs
so
that
our
class
file
transformers
will
be
registered
and
then,
when
they
store
all
the
classes,
when
they
compile
all
the
classes,
they
will
go
through
those
class
file
and
get
transformed.
A
B
Job
agent,
it
would
work
yeah,
except
that
the
they're
gonna
do
tests.
Then
yeah
yeah
yeah,
because.
B
A
A
A
I
think
our
agent
is
complicated
as
far
as
instrumentation,
but
as
far
as
agents,
though
it's
not
that
complicated
right.
We
just
intercept
methods
like
I
can
imagine
a
profiling
agent.
I
don't
know
how
those
are
implemented,
but
that
sounds
much
worse
than
ours,
so
it
might
actually
not
be
that
far.
This
big.
A
C
B
C
C
B
Yeah,
I
was
surprised
I
mean
I
guess
from
a
size
perspective
that
makes
sense,
but
it
seems
like
I
would
have
guessed
that
you
would
just
take
all
the
classes
on
the
class
path
and
compile
them.
A
A
A
A
C
B
I
asked
for
I
asked
for
a
link
where
we
could
follow
a
lot,
follow
that
work
and
he's
like
it's
just
a
bunch
of
internal
hacking
at
red
hat.
At
this
point
it
sounds
like.
B
B
Oh
jason
was
just
giving
an
update
on
the
overhead
testing
and
he
wanted
to
know
where
to
check
in
the
test
results
on
the
nightly
run
and
he
was
concerned
if
he
checked
it
into
maine
that
it
would
trigger
a
build
which
we
looked
up.
I
think
you
found
today
already
there's
a
way
to
exclude
directories
from
triggering
the
build
yep
yep.
You
got
it
cool.
A
Anyways,
I
think
the
workflow
push
doesn't
trigger
the
build
because
they
don't
want
recursive
builds,
so
they
just
don't
without
it.
Oh
what?
If
I
remember
quickly,
I
read
that
maybe
it's
only
the
same
workflow
though
it
might
just
be
the
same,
a
true
recursive
or
not
a
different
word
yeah.
I
know.
Github
has
sun
protection
against
major.
A
C
B
A
C
C
A
A
Incidentally,
internally
we're
having
a
discussion
on
how
to
add
overhead
tests
to
various
repos.
I
think,
starting
with
python
for
the
aws
distro,
which,
with
the
intent
of
hopefully
upstreaming
the
approaches
as
well,
and
there
we're
sort
of
thinking
like
even
if
it
was
upstream,
we
would
use
the
aws
account
to
actually
publish
the
metrics
to
cloudwatch.
So
you
can
have
the
graph
there,
along
with
alarms
and
filters
and
whatnot
have
all
the
features
there.
A
A
I've
I'm
still
like
we
still
haven't
gotten
to
the
point
where
we
know
how
those
would
then
be
published
to
users
which
I,
like,
I
think,
aw
supports
anonymous
dashboards
in
some
way.
Now,
I'm
not
sure,
but
without.
D
A
I
don't
like
to
click
click
sites
or
something
they
have
so
many
features
and
you
can
mix
and
match
them.
So
one
of
those
features,
I
think,
allows
it,
but
I
stopped
following
up
on
that,
but
we're
we've
just
been
talking
about
that.
So
if
that
approach
a
place
we'd,
probably
we
might
consider
it
for
jason's
stuff
as
well
at
some
point
not
for
now,
though,.
C
B
Was
pretty
small
compared
when
nikita
published
the.
A
A
C
C
A
B
Oh,
who
can
we
see
yeah.
C
C
B
B
B
Cool,
oh
and
did
you
I
and
if
you
didn't
see,
we
got
the
github
org
finally,
finally
upgraded
to
enterprise
so
seriously
we
have
180
concurrent
runners
now
from
50.
yeah.
B
Limit
I'm
sure
we
can
break
it,
I'm
getting
prepared.
Now
I
I
started,
I
wrote
something
to
like
use
the
rest
api
to
scrape,
and
here
let
me
share
my
data.
Do
I
have
oh,
I
didn't
save.
Let's
see
how
do
I
well,
maybe
I
can
rerun
it
now.
C
B
Yeah
so
like
I
calculated
the
queue
time
so
how
long
each
of
the
builds
are
sitting
in
queue
and
and
also
the
total
worker
usage.
So
we
can
see
which
repos
are
the
the
culprits.
B
Yeah,
so
I
I
will
have
data
soon,
the
the
darn,
the
the
rate
limiter,
keeps
sporting
me,
though,
so
I
had.
D
A
B
But
I
wrote
because
it
was
annoying
me
so
much
I.
I
now
have
a
little
caching
h2
cache
to
cache
the
the
body
of
the
etag
for
the
uri,
so
I
could
keep
re-running
it
and
supposedly
that's
not
supposed
to
use
your
quota
if
it
if
it
gives
you
back
a
304,
not
modified,
but
we'll
see.
B
A
B
At
which
point
I
like
that's
like
20
lines
of
jdbc,
okay,
yeah,
yeah
cool,
so
all.
A
Right,
I
did
want
to
make
sure
you
talk
about
the
attribute
sampler
yeah,
when
I
saw
that
like
when
nikita
seemed
to
imply
that
this
would
live
in
country
forever.
I
was
very
surprised
because
it
seems
like
a
very
important
sampler
and
another
way
to
look
at
it
is
that
I
wouldn't
expect
our
main
job
instrumentation
report
to
depend
on
contrib,
maybe
the
other
way
around,
but
this
sample
I'm
sure
we
would
use
in
the
java
agent.
A
A
A
C
Okay,
I'm
glad
we're
on
the
same
page.
There
yeah
I
mean
nikita,
I
think,
there's
an
issue.
There
is
an
issue
in
the
spec
rebo
to
track
making
it
beneficial
there's.
You
know
the
other
thing
that
we
that
someone
needs
to
pick
up
and
I
don't
even
know
who,
on
w3c
is
figuring
out
baggage,
metadata
and
actually
getting
getting
in
in
process
only
baggage
that
can
be
used
for
all
sorts
of
stuff,
but
you
mean
that's
going
to.
C
A
Sounds
like
goes
a
bigger
deal
for
the
internal
stuff
than
their
cloud
customers.
That's
my
sort
of
impression.
Oh,
that
could,
and
I
have
a
sort
of
feeling
that
open
time,
which
is
more
of
a
for
the
external
stuff
for
now
yeah,
that
might
be
the
reason
yeah
speaking
of
baggage,
so
yeah.
I
did
want
to
check
with
your
trust
if
you
had
any
thoughts
on
my
comment
about
whether
the
context
key
stuff
is
a
more
generic
problem
that
we
have
to
solve.
A
B
A
B
B
Oh
no,
this
is
like
arbitrary
id.
This
is
my
my
use
case.
At
least.
Is
it's
not
going
to
be
populated
automatically
it's
the
user?
A
B
And
they
I
need
to
get,
I
want
to
give
them.
They
want
to
be
able
to
stamp
that
tenant
id
on
that
on
the
request
span,
which
is
easy,
but
then
they
also
want
that.
D
A
B
B
They
have
to
call
set
attrib
set
attribute
on
the
parent
span
and
they
have
to
set
it
into
the
baggage,
and
then
they
have
to
do
make
current
on
the
baggage
they
have
to
deal
with
contexts,
whereas
the
thing
that
I
like
about
the
just
telling
them
hey,
set,
call
set
parent
on
the
or
set
attribute
on
the
parent,
that's
all
they
have
to
do,
and
then
they
can
give
them
a
configuration
file
where
they
can
say
the
these.
Are
these
two
attributes
I
want
to
prop
it.
C
A
D
C
B
A
A
B
A
A
C
A
C
C
Yeah,
I
remember,
went
to
a
talk
10
years
ago.
I
think
by
one
of
the
one
of
the
founders
of
the
zul
systems,
where
they
had
their.
You
know
like
four
microsecond
garbage
collection
on
like
16
gigabyte
heaps
and
just
like
at
the
time
it
was
just
like
wow,
that's
insane,
but
yeah.
They
had
some
really
crazy,
gc's
well,
and
then
I
don't
know
how
much
of
that
had
to
run
on
their
hardware,
though,
because
like
they
had
that
custom
chip,
the
like
the
custom
chip.
D
C
B
C
Of
kind
of
crazy,
I
mean,
I
think
it
was.
It
was
a
cool
idea,
especially
if
you
wanted
to
run
some
big
big
java
iron
and
like
back
in
the
day,
you
know
back
10
15
years
ago.
You
know
17
000
instances
of
web
logic,
probably.
C
Or
running
yeah,
some
crazy
thing,
yeah
cliff
cliff
click
is
the
was
the
guy
as
a
super
super
sharp
dude
yeah.
He
also
implemented
most
of
the
first
version
of
hotspot,
so
he
knows
his
stuff.
C
What
was
christian's
comments
on
this?
I
don't
remember.
C
A
A
A
Like
I
forgot
about
that,
so
that's
a
great
point
I
mean
so
technically
you
should
have
getters
for
everything.
Then
right
is
that
sort
of
fair?
I
think
that's.
Basically
what
christian
is
saying
it's
like
if
we're
gonna
have
this
together,
we
should
just
like
there's
no
reason
not
to
have
gutters
for
everything,
and
that
seems
okay
to
me.
C
A
A
C
C
C
B
B
Nikita
got
me
wondering
also
about
like
why
do
we
have
certain
in
certain
tests
we
disable
it,
but
it's
already
disabled
by
default.
So
I
was
just
curious,
yeah
and
but
apparently
there's.
Oh
hey.
It
almost
passed
this
time,
yeah
yeah
I
figured
may
as
well
put
these
in
and
at
least
we
have
a
baseline,
then
of
we
know
which
ones
fail.
Although
I
mean
we'll
have
to
watch
out
for
sporadic
failures
for
a
while
yeah
yeah.
D
D
A
B
B
Oh
yeah
yeah.
I
had
opened
this
issue
a
long
time
about
ago
over
a
year
ago,
but
yeah.
B
D
B
B
Oh
okay,
yeah
yeah,
and
then
the
other
point
that
I
think
I
saw
just
a
comment
about
was
that
you
should
cash
the
call
to
this
if
you're
going
to
use
it
in
a
helper,
because
it's
a
little
bit.
D
B
This
implementation
isn't
efficient,
yeah.