►
From YouTube: 2021-10-19 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).
B
B
So
it's
pretty
much
ready
to
go
except
so
I
guess
we
just
need
to
decide
if
we
want
to
try
to
do
this
in
this
release
or
not.
B
I'm
sure
this
pr
is
going
to
you
know,
break
the
release
process
somehow,
but
it's
going
to
break
it
whenever
it
goes
it's
in.
So
I
do
think
it's
a.
I
mean
it's
sort
of
important.
B
If
we
don't
do
this,
the
other
option
is
to
just
manually
bump
and
publish
the
gradle
plugins
right.
The
problem
right
now
is
that
we're
on
a
snapshot
for
gradle
plugins,
so
at
minimum
before
we
make
the
release
we,
we
should
bump
this
and
publish
official
version,
because
that
was
one
of
the
problems
with
the
1
6
release
and
patching
the
1
6
release
was
that
we
had
made
it
on
top
of
a
snapshot
for
gradle
plugins.
B
So
we'll
have
to
look
at
it
now
I
mean,
but
if
you
can
just
take
a
look
at
it
and
basically
let
me
know
if
it
looks
like
worth
trying
for
this
release
or
you
would
rather
hold
off
on
this
release.
D
B
Our
examples
use
don't
use
published
artifacts
at
all.
I
mean
they
right.
They
they
depend
on
the
snapshot
versions
of
all
our
all
of
our
artifacts,
our
instrumentation
api.
B
C
B
Yeah,
so
they
do
for
the
normal
artifacts
and
that's
what
this
crazy
thing
does
right.
We
add
we
published
the
local
snapshots
and
then,
during
the
build
we
substitute
sona
type
for
maven
local.
D
B
If
you,
if
you
go
through
all
the
iterations
of
this,
so
I
tried
a
variety
of
things
and
there
were
a
bunch
of
problems
primarily
because
the
version
I
couldn't
get
nebula
to
work
because
gradle
plugins
is
not
at
the
root
level
and
so
there's
no
version,
so
the
gradle
so
doing
a
publish
to
local.
B
D
D
B
No,
that
that
would
be
fine.
It's
the
reason
why
I
couldn't
get
that
to
work
was
because
it
required.
Let's
look
at
the
changes
along
the
way,
so
you
can
see.
This
is
what
I
I
was
a
midpoint
in
this
pr
was
doing
exactly
that.
It
was
locally
publishing
the
gradle
plugins
to
consume.
Okay,
the
problem.
The
whole
problem
is
around
this
version.
D
B
To
work
because
it
has
no
version,
but
this
was
honorable's
proposal
for
solving
this,
was
we
actually
have
the
version
passed
in
to
the
build
for
the
release
builds
right?
It's
a.
We
have
a
as
a
yaml
pram
yeah,
but
we
could.
B
B
D
A
B
Make
sense
so
do
you
do
you
want
to
try
to
battle
net?
I
mean
it
feels
like
if
there
was
a
way
to
get
nebula
to
work
for
granted.
D
D
C
D
B
B
If
you
look
at
the
updated
sort
of
the
release
instructions,
what
it
would
be
required,
because
we
still
have
a
version
for
the
gradle
plugins
hardcoded
in
there.
D
B
E
D
D
E
E
C
E
D
Well,
what
do
you
mean?
We
cannot
debug
that.
D
D
B
C
D
Yeah,
I
would
say
that
let's
release
1.7,
as
is
even
probably
without
the
grendel
plug-in
version,
and
then
we
can
properly
attack
this.
These
inconsistent
tags
task,
which
will
take
care
of
versions,
thinking
like
more
or
less
automatically.
B
Cool
and
then
once
that
is
published,
then.
B
All
right
we
had
talked
about
this
before
and
I
think
one
of
us
had
even
emailed
the
gradle
enterprise
folks.
A
D
I
well,
I
can't
try
to
write
the
direct
to
hans
he
interviewed
me
a
year
ago,
and
it
seemed
that
he
remembered
me
from
the
days
past
conferences.
D
B
This
this
came
up.
Laurie
brought
this
up.
B
Cool
yeah,
I
would
I
just
I
went
to
this
again
and
I
was
just
surprised
that
they
don't
have
like
anything
more
than
contact
us.
D
C
Yeah
so
during
my
refactoring
of
neti
client
instrumentation
to
instrumental
api,
I
found
some
tests
the
test
time
of
tests
that
have
failed
when
I
switched
over
to
the
net
client
attributes
extractor.
That's
because
the
client
attributes
net.
A
C
Attributes
extractor
takes
the
remote
address
from
the
channel
at
the
end
of
processing
and
when
you
time
out
in
nettie
and
nettie
apparently
considers
the
whole
thing
as
having
failed.
So
it
doesn't
return
any
others,
because
it
fails
to
connect
to
anything.
So
we
don't
get
those
additional
net
span
attributes
this
way
and
yeah.
Besides
it
being
a
regression.
It
very
raised
a
fair
point
that
we
actually
use
some
data
in
a
scenario
that
probably
needs
them
the
most
right.
C
So
the
failed
connection
or
error
response,
so
we
should
probably
rethink
how
to
approach
those
net
attribute
extractors
because,
as
it
turns
out,
there
apparently
may
be
some
clients
neti
based
ones.
I
suppose
that
will
only
expose
the
connection
net
parameters,
everything
on
in
case
of
unsuccessful
scenarios,
of
course,.
C
C
There's
there's
still
one
kind
of
spam:
that's
not
refactored
in
18,
these
are
the
connect
spans
and
I
left
them
for
for
the
very
end.
C
There
will
be
two
more
of
them:
okay,
yeah.
So
when
you
click
on
the
create
instrumenter,
hdp
instrumenter,
you
can
see
that
I'm
using
something
called
neddy
common
net
attributes,
extractors
x,
factor
line
35,
which
actually
extends
the
server
one,
and
I
had
to
command
out
the
p
attributes
extractors,
because
it
was
only
working
with
the
client.
B
C
B
Attributes
extractor
did
it,
does
it
on
both
on
start
and
on
and
right
right
now.
We
only
have
it
on
on
end.
If
we
added
on
start
also
yeah
and
for
neddy,
could
then
use
onstart
if
it
wanted.
B
C
Yeah,
so
I
was
thinking
of
maybe
just
renaming
those
classes,
I
mean,
maybe
if
we
just
get
rid
of
those
fine
server
names
and
call
them.
I
don't
know,
request
response
start
end.
It
will
be
a
bit
a
bit
better,
especially
that
I've
noticed
in
your
pr
trust
when
you
actually
introduce
them.
There
were
several
clients
that
had
all
the
connection
information
available
right
on
the
start,
like
all
the
db,
clients
actually
captured
the
connection
for
on
start.
B
In
the
neti
case,
does
it
work
to
only
capture
onstart,
you
know,
or
do
we
also
need
it
on
and.
B
C
Our
tests
have
passed,
which
is
probably
not
you
know
much
of
an
indicator
about
that.
To
be
honest,
honest,
I
don't
know
really.
I
will
have
to
check
that
if
it's
possible,
it
probably
is
it
kind.
It
depends
when,
when
is
the
connection
phase
executing,
is
it
inside
the
http
request
call
or
is
it
before
the
http
request?
If
it's
before,
then
you
will
have
a
completely
resolved
address
in
the
channel
by
then.
C
If
it's
during
the
http
request,
like
the
first
thing
then
then
yeah,
then
it
would
probably
make
sense
to
try
to
extract
at
least
the
appear.
The
next
hostname
at
at
the
end.
C
E
I
mean
connections
are
always
very
annoying
like
a
client
might
retry
or
something
or
it
could
be
load
bounced
or
something
so
like
just
the
what
the
meaning
of
the
netgear
name.
It's
not
that
clear
anyways
like
that
for
a
very
low
level
thing
like
nettie,
it's
actually
more
obvious
because
of
course,
it's
trying
to
make
a
connection,
but
other
clients
based
on
it
might
still
be
doing
something
else.
C
C
C
B
C
C
E
C
C
E
B
E
C
Don't
think
I
don't
know
if
I
have
enough
net
context,
but
I
looked
up
the
javadoc
on
the
remote
address
method
and
it
says
that
it
will
return
now
if
the
channel
is
not
connected.
So
I
assume
that
when
it
times
out
that
it
treats
as
connection
lost,
then
it's
not
connected
and
that's
why
it
returns
now.
So
it
kind
of
makes
sense.
I
mean
it's,
it's
consistent
with
what
javadoc
says
at
least.
C
Yeah,
so
the
other
idea
that
I
had
was
just
to
remove
the
assertions
on
the
net
attributes
from
from
this
test
case,
which
may
or
may
not
be
captured.
Who
knows,
depends
on
the
library
and
enable
that
in
more
places
so
that
we
at
least
have
in
the
test
case
that
checks
that
we
capture
a
span
for
a
timeout
at
all
in
most
of
our
http
clients.
Because,
even
if
you
don't
capture
the
net
attributes,
you
we
capture
the
http
url,
which
is
like
half
of
the
attributes
anyway.
C
So
just
verifying
that
we
capture
the
http
url
in
case
of
the
timeout
is
probably
a
good
thing
in
all.
B
Yeah
I
liked
I
mean
after
starting
with
this
initially
I
did
like
the
client
server
split.
It
worked
really
nicely
with
the
peer
services.
B
B
And
I
feel
like
we're
other.
This
is
matching
other
things
that,
like
the
http
client,
http
server
split
out
and
like
even
the
span
extractor
kind
of
stuff
we're
seeing
this
client
server
split.
So.
E
C
E
Do
that
I
think
we
could
still,
I
can
on
start
of
the
nutty
instrumentation
we
could
fill
in
the
net
stuff
like
this,
does
a
special
many
code
right,
so
maybe
to
get
those
attributes,
that's
the
simplest
way
yeah.
Let's
just
say
we
document
the
like
yeah
and
he
doesn't
actually
preserve
this
throughout
the
request
response.
Unfortunately,
so
instead
we
record
now:
okay.
B
Cool
it
would
be
really
nice
to
see
how
many
other
http
clients
that
passes
for
causes
problems
for
that's
true,.
B
E
I
wanted
to
comment
about
the
context
performance
thing
you
mentioned
press
yeah
before
doing
anything
agent
specific.
We
should
first
combine
all
our
values
into
one
object
that
we
store
in
the
context.
I
think
we've
discussed
that
a
while
ago,
like
the
first
instrumentation
to
theoretically
only
be
updating
one
object,
anyways
or
something
like
that.
Maybe
two
I
don't
know
depending
on
the
circumstances,
but
not
one
per
field.
E
E
C
B
C
The
server
that
utilities
have
a
few
like
there's
the
one
for
app
server
bridge
for
server,
naming
or
sort
of
that
something.
E
B
Yeah
and
they
kind
of
they
kind
of
cr-
have
different
life
cycles
of
like
some
of
the
netty
ones
were
split
out.
Although
those
were
materials
I
think
were
being
stuck
into
the
netty
connection,
those
were
attribute
keys.
C
C
So
could
we
maybe
group
those
keys
into
not
exactly
one
but
several
classes
like
we
would
have
one
class
for
all
spam
keys.
We
could
have
another
class
for
servered
server.
Then
server
related
things
which
are
usually
used
together
anyway,
and
I
don't
know,
do
we
have
any
other
significant
group
of
keys?
I
don't
think
so.
E
E
Like
I
think,
even
for
the
maybe
just
for
fast
thread,
local,
what
it
does
it
like
in
the
static
initializer,
it
assigns
an
index
to
that
key
and
then
it
just
references
into
the
array
directly
with
context
case.
Maybe
we
have
too
many
contexts
used
in
different
scenarios,
though,
like
the
context
could
could
actually
pre-allocate
its
own
index
and
then
context
is
a
sort
of
a
always.
Has
this
fixed
size
of
all
the
context
he's
used
by
an
app
and
instead
of
having
a
different
size
for
each
context.
B
E
E
E
I
would
probably
avoid
an
agent
first
optimization,
though,
because
I'm
sure
there's
stuff
to
do
before
the
agent
and
then,
after
that
we
could
still
do
some
aging,
because
I
I
thought
your
field
suggestion
was
probably
agent
specific
right
dress,
but
hopefully
you
can
do
something
before
that.
That's
non-agent-specific.
B
Yeah
I
mean,
potentially
we
could
do
that
in
the
even
that
in
the
sdk
context.
Having
the
problem
was
the
context
being
pure
exactly.
E
E
E
E
E
E
E
C
B
E
E
Yeah
it's
much
faster,
I
mean
I
couldn't
get
performance
in
my
benchmarks,
even
doing
checking
both
the
beginning
and
then
like.
I
tried
everything
and
it
was
still
always
faster
to
use
the
array.
Then
that
thing
whatever
that
was.
There
might
still
be
a
different
approach,
though,
that,
like
we
can't
ever
forget
cash
and
like
having
everything
on
the
cash
line,
is
always
going
to
be
pretty
awesome.
Anyways
locations
versus
look
up
speed,
it's
hard
to
know
which
one
is
going.
B
Did
we
have
any?
How
are
you
doing
james.
F
I'm
good
thanks,
I'm
just
kind
of
listening
to
you
guys.
I
don't
really
have
anything
to
to
say
today,
just
kind
of
working
on
my
code
and
listening
to
what
you
guys
are
chatting
about
in
the
background.
F
Yeah,
I'm
getting
right
into
the
depths
of
testing
it
right
now,
I'm
trying
to
record
it
record
like
the
maximum
value
of
a
double
and
the
minimum
value
of
a
double
and
make
sure
it
doesn't
break
and
trying
to
just
push
its
limits
kind
of
so
it's
getting
there.
I
think
it'll
be
probably
ready
to
come
out
of
draft
pr
in
the
next
few
days.
F
So
yeah
it's
gone
going
nicely.
B
B
B
Jason's
yeah,
oh
yeah,
I
think
yeah,
that's
one
this
one
yeah!
So
let's
solve
it.
B
Right
but
this
seemed
fine,
I
didn't.
I
was
thinking
two
different
subclasses,
but
this
makes
sense
to
me
also.
E
B
Yeah
yeah
team
team,
splunk
gotta
gotta,
give
you
guys
some
some
props
there
for.
Thank
you
for.
D
B
B
B
C
E
D
D
D
E
C
D
D
E
F
B
And
how
do
we
bridge
dewey
bridge
with
do
we
support
sleuth,
slash
this
future
or
or
just
say,
that's
springs
like
I
think
to
nikita's
point:
hey
now,
that's
spring
supports
hotel,
it
should
work,
and
but
the
question
is
who,
if
it
doesn't,
who
fixes
that.
D
C
D
D
C
B
E
E
D
E
D
D
F
C
You
would
probably
want
to
have
a
micrometer
to
alter
in
your
java
agent
and
altered
micrometer,
if,
in
your
library,
probably
maybe.
B
Yeah
honorable
dashed
all
my
hopes
like
two
weeks
ago,
when
he
told
me
that
pr
was
going
in
the
direction
that
I
didn't
want.
E
E
D
D
B
Not
for
work,
I've
never
worked
somewhere
where
they've,
given
us
max.
F
I'll
just
be
happy
if
the
new
mac,
my
mac,
when
I
plug
my
screen
in
it,
just
turns
off
the
whole
system
crashes
and
it
restarts
itself.
So
I'm
just
hoping
I'll
fix
that.