►
From YouTube: 2022-09-29 meeting
Description
cncf-opentelemetry@cncf.io's Personal Meeting Room
A
B
B
B
Before
everyone
gets
here
so
I
I
have
that
draft
PR
open
in
instrumentation
that
you
know
updates
to
1.19.0
snapshot
and
makes
all
the
log
changes
do.
Do
we
need
the
kind
of
internal
open,
Telemetry
instrumentation
the
thing
that
you
know
provides
compatibility
between
different
versions
of
the
API.
Do
we
need
that
for
the
for
this
initial
version
of
the
log
API,
or
do
we
only
need
that
once
there's
changes.
C
We
do
need
it
from
the
beginning,
but
it
doesn't
necessarily
have
to
be
at
the
same
time.
Okay,.
C
That?
Because
that
would
only
be
for
people
who
are
using
both
the
agent
and
the
API
on
top
of
the
agent
gotcha.
B
Well,
I
included
that,
but
I
just
wasn't
sure
about
that
architecture.
B
C
C
C
Yeah
I'm
very
I'm,
well
it'll,
be
it's
great
to
delete
that
stuff,
but
I'm
probably
more
excited
about
the
event
API
from
like
an
end.
Use
like
this
is
something
that
we've
been
missing.
B
Like
John
says,
I'm
not
super
thrilled
about
the
ergonomics
of
emitting
events
from
something
called
lager,
but
you
know
I
hope
that
if,
if
users
find
that
confusing,
they
surface
it
up
to
us
and
we
can
use
that
as
a
data
point
to
drive
change.
C
C
So
in
the
instrumentation
repo
to
start
with,
we
kind
of
it's
a
little
wonky.
We
specifically
Set,
have
it
scan
just
a
few
directories
because
we
want
it.
We
want
to
avoid
the
all
the
instrumentation
modules
themselves
since
we've
pretty
much
intentionally
test
against
old
versions
there
and
then
for
suppressing
things.
We
had
a
Jack
was
started
using,
let's
hop
over
to
the
Java
repo.
C
C
And
sort
of
paper
trail
perspective
show
you
what
I
kind
of
found
from
the
this
is
kind
of
interesting.
If
we
go
to
insights
dependency
graph
depend
a
bot
you
can
see
over
here
and
actually,
let's
pop
back
to
the
instrumentation
repo.
C
So,
let's
look
at
this
one
oops
last
check.
We
can
see
the
log
and
it
will
show
you
the
log
file,
and
so
you
can
check
ignore
so
it'll
tell
you,
you
know,
like
slf4j,
simple
was
ignored
from
the
dependabot
ignore
command
these
versions,
which
is
I,
mean
a
way
but
I
still
like
the
idea
of
more
explicitly
doing
that,
and
then
you
can
also
you
can
manually
if
you
ever
want
check
for
updates
here.
C
C
I
think
it's
a
pretty
big,
like
Service,
now
like
of
GitHub
itself,
the
number
of
repos
that
it
must
be
scanning
and
managing.
D
C
Yeah,
we
were
not
using
it
up
until
a
couple
of
weeks
ago.
Okay,.
A
It
was
one
just
one
thing
to
note:
I
I,
remember,
reading
something
about
it
is
if
you
Fork
a
repository
that
has
to
pen
about
depend
about
enable
your
fork
might
get
it
as
well.
So
you
might
get
sort
of
those
PR's
in
your
repository
and
you
end
up
with
clutches
I
think
that
was
some
critique
I
remember.
Reading
about
it,
I've
been
using
renovate,
which
is
a
similar
tool,
but
it's
not
integrated
as
much
as
dependable
yeah
like
there's
something
to
consider.
If
we
start
hitting
issues
like
that.
C
Yeah
this
here's,
my
Fork,
yes,
I-
have
been
getting
them
in
my
Fork.
B
Have
you
figured
out
how
to
turn
those
off
I
have
a
fork
of
The
Collector
and
it
The
Collector
obviously
has
a
lot
of
dependencies
or
collector
control,
and
so
it's
like
constantly
spammed
with
depend
about
PR's
and
I,
tried
well,
I,
didn't
try
everything
I
could
think
of,
but
I
spent
like
15
minutes
trying
to
mess
with
the
settings
and
deactivate
actions
and
things
like
that.
C
Yeah
I
I
found
out
this
recently
look
at
6
000
depend
about
PRS.
A
If,
if
it's
a
minor
patch
like
if
you're
updating
a
minor
dependency,
it
can
do
it
without
a
PR.
Even
it
will
sort
of
wait
for
the
test
to
finish
and
then
just
merge
it
straight
in
so
you
can
reduce
the
noise
with
some
of
those
things.
There
might
be
just
something
to
consider.
I,
don't
know
how
how
you
would
run
it
on
on
sort
of.
We
could
probably
run
Nintendo
circles
the
eye
or
which
the
build
pipeline
we're
using
unless
they
have
a
different.
C
A
Like
yeah,
it
has
started
some
of
it
is
out
of
the
box.
If
you
take
their
default,
configuration
almost
right
and
you
can
even
say
like
they're
all
miners
in
one
go,
so
you
avoid
a
little
bit
of
that.
So
if
it
takes
three
hours
to
you
approve
everything.
You
know
that
it
takes
four
hours
for
everything
to
merge
with
the
rebates.
C
Yeah,
do
you
know
if
I
renovate
has
is
free
for
open
source.
C
Yeah
I'll
take
a
peek
at
it.
I
know,
I
mean
dependabot
is
generally
what
other
Hotel
repos
are
using,
but
it
would
be.
There
are
some
downside.
Stuff.
Fork
thing
is
annoying,
though
yeah.
B
I
found
the
issue
on
GitHub
about
it,
and
it's
there's
a
comment
from
well.
I'll
include
it
in
the
notes
first,
but
there's
a
comment
from
the
maintainers
that
depend
about
that
said:
they're
working
on
a
fix
for
it.
That's
two
years
old.
It's
it's
a
it's
a
massively
popular
issue.
C
Yeah
I
think
they
I
read
another
issue
over
here
about
cradle
support,
which
has
which
was
very
popular,
or
maybe
it
tied
to
some
other
issue
that
was
very
popular
and
the
maintainers
mentioned
that
they've
just
been
swamped
with
scaling
the
service
out,
since
they
got
subsumed
by
GitHub
yeah.
F
C
E
And
I
I
mean
it's
so
it
seems
like
no
one
else
cares
about
this
and
you
know
a
user
that
is
able
to
control
their
environment
shouldn't
get
themselves
into
such
a
predicament.
However,
it
does
seem
like
bad
behavior
to
have
to
have
there
be
because
these
are
kind
of
map
Fields
right.
They
contain
multiple
key
values
to
have
one
stump
over
the
other,
where
it's
absent
seems
like
a
mistake.
So
that's
what
this
was
hoping
to
address
in
a
very
surgical
kind
of
small
way.
E
E
Can
I
can
talk
a
little
more
about
it,
so
yeah,
specifically
if
the
environment
variable
contained
both
the
service
name
and
the
environment,
for
example,
and
the
system
property
contained
only
the
service
name.
The
absence
of
the
deployment
environment
in
the
system
property
would
then
cause
it
to
be
emptied
out
right.
It
would
be
cleared
and
set
to
null
and
there
would
no
longer
be
a
deployment
environment
set,
which
is
probably
not
what
the
user
wants
and
in
fact
the
spec
does
say
like
if
user
provided
stuff
should
be
merged.
E
E
E
B
I
was
trying
to
think
of
it
from
like
a
like
a
framework
perspective
like
if
I
was
using
spring
and
I
had
a
application
properties
file
where
I
specified
like
the
port
or
any
of
the
other
properties,
and
then
I
also
specified
a
like
a
command
line
argument
or
a
system
property
I.
You
know,
I
I
wouldn't
expect
those
to
be
kind
of
merged
together.
B
I
would
expect
one
to
take
precedent
over
the
other,
and
you
know
that's
kind
of
the
the
opinion
we've
taken
here
is
that
system
properties,
Trump
environment
variables
and
that's
the
only
reasonable
thing
you
can
do
for
all
the
different
types
of
properties
that
are
configurable,
except
for
maps
like
any
of
The
Primitives,
whether
they're,
like
numbers
or
strings
or
durations,
there's
no
kind
of
useful
way
to
merge.
Those
same
thing
goes
for
list
lists
or
raise
because
you
know
you
can't
just
concatenate
they
arrays
together.
B
The
ordering
is
gonna
is
gonna
matter
in
there.
Maps
are
the
only
thing
that
this
is
really
a
question
for
yeah
and
I
yeah
I,
don't
know
like
I'm,
not
I
I
thought
we
should
keep
the
behavior
for
maps
the
same
as
the
rest
of
the
configurable
types
for
consistency's
sake.
B
At
the
same
time,
I
don't
think
it
would
be
the
end
of
the
world
if
we
made
this
change
because
I
looked
at
the
usage
of
you
know,
map
based
configurable
properties
and
I.
Think
there's
just
this
and
one
other
type
that
used
Maps.
So
you
know
it's
not
something
that
happens
very
often.
C
B
Then
obviously,
you
know
the
additional
thing
is
that
you've
done
this
in
a
surgical
way,
so
it
only
applies
to
this
one
property.
I,
don't
know
if
that's
a
good
thing
or
a
bad
thing,
I
think
if
we
were
to
do
this,
we'd
I'd
make
it
generally
applicable
to
map
types
not
have
like
a
special
kind
of
asterisk.
Around
resources.
C
I'm
not
sure
about
all
maps
because,
like
you
need
some
context
like,
is
this
a
map
of
independent
things
in
it?
Or
is
this
a
map
of
like
say
an
instrumentation
I
want
to
make
a
a
dynamic
instrumentation
at
this
class
name,
this
method
name,
this
signature.
B
And
that
and
that's
kind
of
where
actually
I
landed
at
that's,
why
I
I
landed
at
let's
take
a
hard
line.
Stance
and
just
be
consistent
everywhere
is
because
I
I
looked
at
the
usage
of
the
map
type,
and
there
was
only
a
couple,
and
so
you
know,
I
was
like
well
I
think
our
Behavior
should
be
consistent
across
you
know
all
maps,
not
just
have
this
special
case
for
resources
and
then
I
kind
of
came
to
the
conclusion.
You
did
Trask
that
well.
B
This
may
make
sense
for
the
maps
map
configuration
properties
that
we're
aware
of
today,
but
we
may
introduce
them
in
the
future
where
this
doesn't
make
sense,
and
we-
and
we
may
regret
that-
and
so,
like
you
know
at
the
conclusion
I
was
I
came
to
was:
let's
just
keep
it
simple
and
consistent,
and
you
know
have
one
I
like
a
priority
system
where
system
properties,
Trump,
environment
variables,
wholesale.
C
Jason
a
question:
if,
if
a
user
like
say
with
the
Java
agent,
we
do
have
that
properties
file
also.
C
E
E
It
might
seem
a
little
bit
contrived,
but
I
think
there
are
users
that
deploy
Java
apps
into
environments,
that
they
don't
necessarily
control
like
they
have
an
operations
or
deployment
team
that
sets
up
the
environment
like
they
like
your
Ops
Team,
might
control
when
environment
variables
are
available
to
your
jvm,
but
then
the
developers
of
that
application
might
have
control
over
what
system
properties
like
what
command
line
arguments
are
on
there,
and
so
it
could
be
a
way
for
two
teams
to
play
nicely
with
each
other
right
if
the
deploy,
if
the,
if
the
deployment
environment
is
owned
by
an
Ops
team
and
they
specify
resource
attributes,
deployment,
environment
and
I.
E
C
Mean
you
know
jamming
you
value
pairs
together,
totally
there's
attributes,
yep
and
so
I'm,
just
trying
to
think
long
term.
You
know
we'll
probably
have
some
kind
of
yaml
configuration
the
app
owner.
Can
you
know,
set
up
their
resource
resource
attributes
in
that
gamma
file,
and
then
this
could
this
would
be
a
different
property
than
could
be
used
to
add
additional
things.
D
Let
me
have
another
perspective,
so
both
on
quarkers
and
spring
and
other
other
Frameworks
configuration
are
handled
by
the
framework
and
in
the
case
of
quarkers,
you
can
specify
configurations
well
variables.
You
can
create
from
files.
Yaml
files
even
can
create
resource
to
grab
them
from
a
database
or
whatever
aren't
we
replacing
ourselves
to
the
framework
providers
where
these
apps
are
being
deployed.
D
So
if
we
create
the
system
where
we
can
load
things
from
files
and
well
eventually
other
places
in
here,
that's
the
duplication
of
work
from
containers
that,
where
people
deploy
these
types
of
applications,
the
only
place
I
see
where
the
Piston
Java,
where
this
would
be
useful,
is
a
plain
Java,
main
method.
Where
you
start
things
on
your
own
and
you
don't
have
any
framework
that
helps
with
configuration
and
you
can
load
these
results
from
a
file
and
that's
fine.
B
B
D
My
my
concern
is,
if
we
create
kind
of
a
situation
where
the
the
Frameworks
are
being
bypassed
and
things
that
they
expect
to
be
there
in
a
certain
way,
are
not.
E
Yeah
I
mean
I
I,
think
I
hear
where
you're
coming
from
like,
if,
if
you're,
if
you're
deploying
spring
or
caucus
or
some
system,
that
already
has
a
configuration
subsystem
shouldn't
we
be
able
to
bootstrap
from
that
like
existing
configuration,
subsystem
I
think
it's
a
little
a
little
bit
detached
from
this
specific
resource.
Attributes
thing
but
I
mean
I.
Think
it's
a
it's
a
reasonable
question
to
be
asking
I.
Think
well.
A
B
And
so
they
they
have
open
Telemetry
configured
in
there,
and
you
know
they
they
do
some
magic
where
they
take
their
configuration
file
for
the
framework
in
general,
extract
some
properties
and
use
it
to
configure
the
open,
Telemetry
SDK,
and
so
that's
nice.
B
So
you
can
use
your
existing
Micronaut
configuration
scheme
to
configure
open,
Telemetry
but
I
think
where
the
auto
configuration
module
comes
in
and
where
the
system
and
environment
variable
based
configuration
schemes
come
into
play
is
like
you
know,
you
should
still
be
able
to
take
advantage
of
that
simple
configuration
setup.
If
you
don't
have
one
of
those
Frameworks
so
like.
If
you
don't
have
spring
and
you
don't
have
Micro
knot
like
you,
should
still
be
able
to.
You
know
quickly
and
simply
configure
the
SDK
without
having
to
do
it.
Programmatically
and
so.
D
D
I
agree,
but
I
I
would
suggest
that
we
create
some
kind
of
obstruction
that
those
Frameworks
can
hook.
F
I
mean
one
of
the
one
of
the
main,
the
primary
tenets
that
I
insisted
on
at
the
beginning
of
open
till
my
open
Telemetry
work
was
that
all
of
the
configuration
must
be
available.
Programmatically
and
Frameworks
can
use
that.
However,
they
like
so
I,
don't
think
that
we
need
anything
else.
If
Frameworks
can
apply
the
configuration
programmatically,
they
can
provide
whatever
configuration
system
they
want
without
any
problem.
I,
don't
I,
don't
feel
like
there's
anything
we're
missing
here.
B
It
does
something
kind
of
in
between
where
it
uses
the
like:
the
auto
configuration,
module
and
kind
of
tries
to
layer.
You
know
configuration
from
environment
variables
and
also
from
their
own
Micronaut
configuration
scheme
and
merge
them
in
some
interesting
way.
It
basically
extracts
the
properties
out
of
their
yaml
scheme
and
then
translates
them
to
new
properties
Source.
B
So
you
know
those
kind
of
get
interwoven
into
the
system,
properties
and
environment
variables
and
there's
a
priority.
It's.
D
Yeah,
that's
kind
of
what
I'm
doing
on
the
work
to
include
the
auto
configuration
on
on
quercus,
because
we
have
the
the
configuration
system
that
already
deals
with
files
and
environment
fireables
and
all
that
and
people
have
just
one
file
where
they
Define.
They
can
Define
all
the
properties
if
they
want
and
we
grab
that
and
inject
those
properties.
On
the
auto
configure.
F
D
Because
the
configuration
brings
a
lot
of
a
lot
of
advantages
and
a
lot
of
work
that
we
don't
have
to
replicate
and
the
only
part
that
we
are
adding
on
top
of
it
is
kind
of
introducing
validations
for
the
properties,
because
we
create
a
model
and
we
can
say:
oh,
you
are
introducing
data
that
doesn't
make
sense
on
Sense
on
this
property
and
we
tell
that
to
the
user.
D
So
we
kind
of
curate
the
properties
ourselves
and
but
then
there's
a
lot
of
things
that
need
to
be
integrated
together
with
the
auto
configuration
because
there's
CDI
and
bins
producing
CDI.
That
needs
to
play
along
with
things
that
are
loaded
by
the
SPI,
and
we
want
to
to
to
also
support
all
the
properties
that
are
on
the
auto
configuration.
So
if
someone
goes
to
the
hotel
page
and
see
a
property
that
we
support,
I
want
that
property
to
be
supported
the
same
way
on
quercus
I.
D
Don't
want
this
to
be
creating
kind
of
parallel
universe
of
configurations
for
us.
F
So
I'm
hearing
two
very
conflicting
things
from
you:
Bruno
one
is
yeah
I
I
I
want
to
do
my
own
configuration
and
I.
Don't
want
to
do
my
own
configuration
like
it
feels
like
you
kind
of
want
your
cake
and
eat
it
too.
I
mean
if
you
want
to
have
a
custom
configuration
thing
for
caucus.
I
mean
you're,
absolutely
welcome
to
when
we
provide
all
the
facilities,
but
I
don't
think
open.
F
D
So
let
me
rephrase
and
simplify
a
bit,
so
what
I
want
to
do
is
kind
of
support.
Whatever
configuration
has
the
Auto
configure
our
own
way
and
with
additional
requirements
that
we
have,
but
apart
from
that,
if
there
is
alternative
ways
on
the
auto
configure
to
load
properties
like
from
files
and
if
that
file
is
present,
we
would
like
to
have
a
way
at
least
to
to
understand.
D
Okay,
this
guy
is
loading
things
from
this
file
and
we
need
to
take
that
into
account
and
if
we
want
to
to
to
have
a
validation
step
or
something
should
have
at
least
who
got
something
to
to
know
that
those
properties
are
being
loaded.
So
we
can
act
if
needed.
Just
that.
C
So
you
want
to
be
able
to
well
do
you
want
to
just
shut
off
like
the
default
property
sources
for
otel
and
just
be
the
have
the
only
control
of
the
property
sources?
Or
are
you
wanting
to
merge
one
hotel
to
read
in
its
properties
via
normal
places,.
D
I
I
I'm
I'm
flexible
in
here,
because
the
way
it's
done
right
now,
it's
it's
good
for
us,
because
we
can
pass
the
properties
for
the
auto
configure
and
it's
what
we
do.
We
create
those
properties
and
we
pass
them,
but
if
you
have
to
configure
also
after
that
loads,
a
file
right
now,
I
don't
have
a
way
to
understand
that
this
file
file
has
been
loaded
and
it
has
any
conflicting
issue.
I.
C
C
B
In
the
class
is
called
Auto
configuration
customizer.
C
Properties,
customizer
yeah,
so
we
had
requested
this
from
the
a
Java
agent
distros,
also
where
we
can
get
the
existing
properties
that
were
loaded
and
we
can
overlay
new.
We
can
basically
calculate
new
things
based
on
the
properties
that
were
provided.
C
Yeah
I
mean
my
my
perspective
is
that
you
know
we
I
mean
if
there's
something
we
can
do
in
the
auto
configure
like
a
an
additional
hook
or
something
to
support
other
configuration
systems.
D
Foreign,
so
something
that
would
work
for
me
is
okay,
this
this
we
are
changing
the
properties
based
on
a
file
or
anything
else,
and
if
we
have
a
hook
that
receives
those
properties
can
receive
some
function
to
filter
things
or
do
any
operation
with
them
and
then
return
the
the
properties
that
I
think
that's
simple
enough
and
it
worked.
It
would
work.
I
guess.
C
Yeah
check
it
out,
it's
relatively
new
in
the
auto
configure
SPI,
okay.
B
And
it
has
a
bit
of
a
caveat,
so
the
properties
that
the
format
that
you
get
properties
and
you
have
the
ability
to
customize
them.
It's
a
it's
a
config
property.
So
it's
not
like
you
get
a
map
of
all
the
properties
and
you
return
a
map
of
all
your
customized
version
of
those
properties
that
have
been
like
filtered
or
whatever
dispensary.
B
E
E
Well,
I,
don't
know
that
we're
any
closer
on
my
PR
I
seem
to
be
the
only
one
that
thinks
that
it's
helpful
and
it
doesn't
seem
like
there's
a
lot
of
support
for
it.
So
I
think
I'm.
Okay,
closing
it,
but
I
do
want
us
to
maybe
just
reread
that
bit
of
the
spec
and
just
make
sure
that
we're
we're
comfortable,
not
merging
in
the
environment
portions,
because.
E
F
It's
a
it's
a
weird
thing,
because
we
have
two
sources
of
user:
a
user
provided
information,
yeah,
so
Jason
Jason.
Just
for
for
my
identification
and
may
actually
be
great
to
ask
anyone
else:
are
there
do
vendors?
Have
customers
who
have
complained
about
this
are
having
troubles
with
it
like
I
guess
is.
C
F
I
mean
I
think
it
might
be
so
I'm,
not
I,
don't
think,
there's
anything
necessarily
wrong
with
this
PR.
Another
approach
could
be
just
that.
We
document
like
if.
E
That
was
going
to
be
my
caveat
like
if
I
think,
if
we
close
this
I
think
maybe
documenting
that,
like,
if
you
specify
it's
stomping
over
and
any
even
like
the
components
in
the
map
of
the
other
thing
will
not
be
carried
forward.
I
think
just
documenting.
It
is
fine.
It
might
be
nice
to
log
a
warning
so
that
if
it
does
happen
in
the
wild,
we're
not
fighting
with
it
and
trying
to
figure
out
what
happened
and
I
mean
the
the
pr
is
hacky
I
mean
come
on.
B
We
before
we
stop
talking
about
this,
here's
one
novel
idea:
okay,
what
what?
If
we
switched?
What?
If
we
added
two
additional
resource
providers,
one
that
configured
a
resource
from
the
config
properties
and
one
that
configured
the
resource
from
the
environment?
So
one
from
you
know
the
actual
Hotel
underscore
resource
underscore
environments,
environment
variable,
and
so
they
just
got
loaded
like
the
other
SPI
based
resource
providers
and
they
got
merged.
Naturally,.
E
A
B
So
you
know
the
advantages:
is
they
get
merged?
Naturally,
we
don't
have
to
change
like
the
definition
of
how
map-based
configuration
properties
work
at
all,
and
you
know
you
actually
gain
some
additional
control
too,
because
you
know
you
have
the
through
other
system
properties.
You
have
the
ability
to
activate
or
deactivate
specific
resource
providers
by
referencing.
E
C
B
B
D
C
At
the
property
at
the
config
property
level,
I
think
everything
is
this:
it's
been
normalized
already.
B
C
B
F
E
C
Yeah,
my
my
overall
thought
is
merging
configurations
is
always
painful
and
there's
rarely
like
a
ideal
solution.
So
I
tend
to
lean
towards
the
simplest
which
would
be
just
ignore.
You
know
these
are
override
each
other.
E
Jack
wallet,
while
we
have
you
your
comment
down
below
here
about
taking
a
Hardline
stance,
you're,
not
suggesting
that
we
shouldn't
support
both
you're,
just
saying
where
they
conflict,
we
shouldn't
have
to
emerge
yeah.
B
E
E
So
I'm
just
adding
some
notes
about
the
other
stuff.
Okay,
so
we
had
a.
We
had
a
case
where
a
user
had
misconfigured
their
otlp
endpoint
or
something,
and
basically
our
our
exporter
got
pointed
it
I
think
it
was
just
like
some
vanilla
web
server
and
the
end
result
of
that
is
a
very
cryptic
message
that
is
hard
to
troubleshoot.
So
it's
like
clear
that
there's
like
some
exporter
problem,
but
it's
really
like
it's
in
the
bowels
of
like
I.
E
E
E
And
so
I
was
wondering
if
it
would
make
sense
or
if,
if
there's
any
consideration
to
maybe
doing
like
a
connection,
setup
ping
and
I,
don't
know
grpc
well
enough
to
say
if
that's
like
a
supported
feature,
I
had
some
evidence
that
it
might
be,
but
just
to
like,
when
the
connection
is
first
set
up
just
even
like
when
the
exporter
is
stood
up.
Is
there
a
way
to
just
to
say
like?
E
F
There's
a
there's,
a
small
chicken
and
egg
issue
approach,
and
then
you
might
be
bringing
up
the
service
before
the
collector
is
ready,
and
you
wouldn't
want
this.
You
would
want
the
service
to
probably
be
running
and
potentially
spewing
errors
yeah
until
the
collector
was
up
and
then
had
it
start
working.
So
there's
a
totally
there's
an
interest.
There's
a
there's.
A
precedence
and
that'll
completely
depend
on
how
the
user
wants
to
their
service
to
behave
like
do.
C
F
We
spent
a
bunch
of
work
doing
that
for
some
class
of
grpc
exceptions,
but
this
is
obviously
one
that
we
did.
We
didn't
get
didn't
see
or
didn't,
find
yeah.
E
No
I
haven't
I,
haven't
really
dove
into
that
part
of
it.
Yet.
But
that's
a
good
idea
because
really
I'm
just
trying
to
make
it
easier
to
troubleshoot
when
this
happens,
because
it
was
not.
C
E
Mean
yeah:
no,
it
would
probably
help
like
support
people
that
are
troubleshooting
customer
problems,
but
I
mean
it
just
depends
on
the
environment.
It's
often
hard
to
get
tools
and
to
secure
environments
or
Docker
environments.
I,
don't
know.
B
Just
extract
some
more
normal
things
from
the
HTTP
request,
because
that's
what
often
happens
right,
you
know
this
request
is
going
through
a
load
balancer
or
something
like
that,
and
you
know
if
the
service
on
the
other
end
of
the
load,
balancer
stops
and
it
stops
being
able
to
respond
to
requests.
The
load
balancer
will
send
like
a
a
503
unavailable,
which
is,
and
it
won't
send
that
in
like
a
grpc
compatible
like
and
so
like.
B
You
know
that
information
it
gets
passed
to
the
the
client
via
just
regular
old
HTTP
in
our
grpc
clients
aren't
able
to
like
make
sense
of
it.
So
you
know
if
I
could
just
know
that
this
there
not
the
grpc
status,
but
the
HTTP
status
was
503
and
try
to
I
don't
know,
maybe
interpret
the
body
of
the
requests
as
a
as
a
string,
yeah
and
print
that
out,
and
that
would
go
pretty
far.
C
I
agree
that
putting
investing
time
into
making
it
easier
to
identify
these
as
well
time
well
spent
I,
know.
On
our
end,
we
don't
use
otlp
grpc,
but
connection
problems
are
like
one
of
the
like
the
most
common
support
issues.
What
do
you
use?
C
So
we
see
a
lot
of
like
SSL
cert
problems,
whether
it's
you
know
running
an
old
jdk
that
doesn't
have
the
latest
search
trust
store
or
they
have
some
Custom
Distribution
that
doesn't
have
the
the
SSL
libraries
yeah.
We
basically
do
this.
You
know
with
catch
those
and
try
to
inspect,
for
what
we
know
are
common
things
and
we're
using
Neti.
So
it's
you
know
some
underlying
Neti
exceptions
that
give
us
Clues
and
then
we
log
something
more
useful.
G
G
G
Well,
I'm,
a
responsible.
It's
always
for
in
I,
can
call
say
for
all
client
apis
for
our
messaging
broker,
open
Telemetry
integration
and
also
I
built
the
first
generation
or
first
prototype
of
our
source
receiver,
which
is
now
a
part
of
the
community
code.
So
a
while
ago,
I
asked
how
we
can
do
as
a
custom
propagation
on
Slack,
and
it
was
a
great
suggestion
using
reflection
because
there
are
a
few
say,
a
shortcomings
with
a
class
loading
and
so
on.
G
But
when
we
try
increase
our
say
customer
and
our
in
our
internal
pocs
to
get
any
kind
of
linking
or
context
propagation
between
publisher,
app
and
Receiver
app
and
Receiver
app
using
Auto
instrumentation
JMS
using
a
synchronous,
receive
we
where
say
I'm,
not
saying
surprised,
but
we
we
noticed
that
there
is
absolutely
no
way
say
for
the
user
to
do
any
kind
of
troubleshooting
and
the
open
Telemetry
is
very,
very
helpful
to
say,
help
user.
G
To
find
say
issue
and
they
have
a
huge
setup,
not
like
a
tiny
app
but
thousands
of
clients
connected
and
then,
for
example,
somebody
published
a
message
and
the
receiver
did
not
receive
the
message
and
or
received
a
message
from
Samsung,
but
it
was
not
intended
to
be
received.
Then
there
is
absolutely
like
no
nice
way
in
the
back
end,
like
Diego
a
Zipkin
to
see
the
trace.
G
So
now,
for
example,
we
managed
to
make
a
seamless
tracing
between
like
get
this
plate
and
Jager
and
Zipkin
between
our
public
or
any
kind
of
publisher
and
Receiver
app
going
through
the
say,
Solace
broker,
including
what
happens
within
a
service
broker
basically,
but
the
receiver
site
is
dark,
so
synchronous
receive
is
now
in
darkness,
and
people
like
user,
basically
or
customer
won't
have
this
ability,
unless
they're
using
mono
instrumentation,
which
we
not
necessarily
encourage
them
to
do.
G
I
have
seen
like
a
bit,
but
I
will
go
PR
on
a
messaging
semantic
convention,
page.
G
Like
about
like
this
is
basically
appear,
it's
about
context,
propagation
in
general,
about
separation
in
creational
and
transport
context,
and
it
was
a
section
with
say
examples
which
suggests
that
making
a.
F
G
G
So
do
you
think
in
Gmail's
Auto
instrumentation?
This
would
be
say
available
addition,
and
if
the
answer
is
yes,
then
how
we
could
get
the
thing
implemented,
because
now
there
is
a
say
receive,
is
disabled
by
default
and
there's
an
experimental
properties
enabling
this
and
then
it
enables
application.
Trace
is
apparent
of
a
receive
Trace
I'm
not
going
to
debate
this.
This
is
going
to
be
a
long
debate
now,
but
in
in
case
say
of
a
process
trace.
It
is
a
message
listener.
G
Where
picture
is
very
clear
who
published
the
message
here,
a
picture
is
blur,
so
you
could
guess
based
on
on
which
topic
it
was
published
and
maybe
some
metadata
which
it
could
be
say,
part
of
the
message
but
still
manual
instrumentation
is
required,
and
this
is
well
I,
don't
see
problem,
but
it
won't
necessarily
contribute
to,
especially
if
use
of
some
older
systems
to
adopt
Telemetry
and
say.
G
C
So
do
you
mean
linking
in
the
sense
of
you
mentioned
that
experimental
option?
Yes,
you
enable
that.
Does
that
give
you
the
linking
that
you're
looking.
G
G
So
basically,
when
I
have
a
method
say
and
get
my
message
and
I'm
using
Auto
instrumentation
only
and
I
I
would
configure
Java
Auto
instrumentation
to
say
on
a
class.
Whatever
my
message,
receiver
give
a
parameter
and
get
it
instrumented
then
app
is
running.
Then
the
parent
of
the
receive
span
is
is
say.
G
Could
be
back
but
as
a
side
as
a
I
had
Impressions
that
there
is
no
propagate
involved
at
all
in
this
floor
right.
So
when
this
there
is
no
propagator
involved,
then
I
don't
know
how
we
could
or
how
API
could
extract
a
trace
parent
to
make
link.
C
C
G
One,
oh
foreign,
so
there
is
like
a
propagator,
is
not
involved
and
when
I
say
register
my
agent
extension
so
I
built
an
extension
that
can
substitute
a
propagator
and
I
build
the
same
composite
propagator.
That
acts
as
a
this
reciprocator
but
has,
for
example,
for
Solace
messages,
has
a
particular
knowledge
of
some
interfaces,
internal
interfaces
and
and
extracts
information
for
any
another.
Jms
message:
it
does
work
as
it
does
right
now,
as
a
VSC
propagated
us
with
internal
get
and
SATA,
but
not
having
a
agent
involved
or
not
providing
a
link.
C
Yeah
yeah
no
I'm
just
wondering
why
it's
I,
don't
remember
why
it's
like
that
because
and
I
was
checking
to
see
if
Kafka,
for
so
the.
C
D
I
think
I
think
there's
the
Gap
in
the
spec
there's
a
messaging
STI
meeting
that
happens
right
before
this
one
and
over
there
they
they
are
discussing
how
to
change
these,
not
just
the
level
of
the
semantic
and
vegetables
also
what
needs
to
be
propagated
and
how?
Because
right
now,
it's
not
clear.
G
So
there
is
a
say:
is
it
this
is
Pierre
I
paste
it
for
the
convention
like
messaging
convention,
and
there
is
a
here
they're
mentioning
batch
message
producer,
but
in
general
like
it
is
any
message
producer
with
internet,
creates
pens,
single
message,
pool
consumer
and
picture
obviously
shows
a
link
on
receive.
A
G
G
Supports
now
a
linking
yes
I,
but
but
but
but
say
I
would
say,
I
don't
want
to
say
this
is
a
problem,
but
this
great
situations
where
it
pushes
user
to
use
certain
interface
and
JMS
they
may
not
always
can
use
when
they
wish
to
have
visibility
in
the
app
like
cross
a
boundary
because
it's
like
making
a
receive
span.
A
child
of
application,
spin
Cuts,
any
kind
of
close
Service
tracing
here.