►
From YouTube: 2023-02-08 meeting
Description
cncf-opentelemetry@cncf.io's Personal Meeting Room
B
D
A
So
I
will
share
and
let's
pull
up
these
Network.
A
Okay,
so
I
think
Alex.
So
the
the
one
thing
that
lydmill
and
I
were
having
some
trouble
with
is
the
the
source
and
destination.
A
So
we
were
wondering
if
it's
kind
of
so
in
open
Telemetry,
we
have
the
those
sock
Pier,
which
is
like
meant
for
the
socket
level.
The
direct
connection
over
the
socket
that
we're
observing
so
is
that
I
was
thinking
that
that
is
the
same
as
your
source
and
destination,
which
seemed
to
be
very
like
a
lower
level
network
connection
versus
client
server,
which
is
what.
B
Yeah
I
I
think
the
the
intention
of
of
source
and
destination
is
a
different
one.
So,
as
I
wrote,
it's
it's
really
from
the
perspective
of
a
packet.
So
if
you
have
a
client
server
connection,
this
stays
all
the
time.
The
client
stays
the
client
and
the
server
the
server
right.
But
if
the
server
responds
to
the
client,
then
from
the
packet
perspective
the
source
and
destination
just
switches,
so
the
source
or
the
packet
is
then
the
server
and
the
destination
is
declined.
Oh.
B
I
think
for
for
the
purposes
what
we
are
discussing
here,
yeah,
the
better
choice,
is
to
to
consider
server
and
client
and
namespaces
yeah.
B
A
Oh
and
one
last
just
for
my
clarification,
then
the
source
and
destination
are
just
the
same
thing
as
client
and
server
just
they
might
be
swapped,
but
the
same
information
goes
in
those
yeah
exactly
got
it.
Thank
you.
E
B
Yeah
I
think
it
really
depends
on
on
the
context
of
usage,
so
ECS
is
not
just
about
observability.
It's
also
used
a
lot
for
security
use
cases
where
you
would
do
like
analysis
of
network
packets
and
I
think
this.
This
is
where
source
and
destination
came
from,
where
you're
really
on
a
packet
basis
would
like
to
know.
B
Where
is
this
packet
coming
from
and
where
is
it
going
to,
while
the
client
server
namespaces
are
really
more
for
the
use
cases
we
are
discussing
here,
where
we
have
a
stable
connection
between
a
client
and
a
server.
E
Yeah
yeah,
it
makes
sense,
but
then
trying
to
understand
is
if
they
are
always
the
same.
It's
probably
not
interesting.
Are
there
cases
where
they're
different,
but
if
I
have
both
of
them
on
my
record,
but
they
are
different.
E
I
mean
the
combination
of
okay.
Let
me
just
try
to
put
it
correctly.
So,
if
I
have
both
what
server
and
destination
right,
their
server
and
source,
which
is
the
right
player
like
what
is
the
what's
possible
to
get
on
the
server
side,
source
and
server
right.
E
B
E
A
Okay,
that
so,
let's
talk
about,
then
the
NAT
attributes
in
yes.
B
A
So
lud
Miller
in
the
in
this
example
here
would
it
be
just
removing
the
gnat
from
all
of
these.
E
So
in
the
first
example,
yes,
but
we
allowed
to
optionally
populate
socket
address
right.
So
if
somebody
wants
to
populate
okay
so
that
yeah,
but
then
it's
an
intermediary
case
yeah.
So
the
direct
connection
is
just
this
without
the
apps,
without
nothing.
B
E
For
Unix
domain
socket
no,
but
definitely
think
further
than
that,
then,
let's
think
about
it.
In
a
Bluetooth,
there
are
some
Android
instrumentations
in
up
until
the
material
world.
Bluetooth
has
port,
but
it
doesn't
have
IP.
B
E
Yeah
I'm
thinking
so
like
I
think
we
in
up
until
I'm
a
train
elastic
common
schema.
We
have
a
similar
approach
where
we
say:
okay,
there
is
this
logical
thing
right
in
your
case,
it's
a
dress.
In
our
case,
it's
its
name
and
then
you
can
understand
what
it
is
by
some
specific
attributes
so
like
if
I
have
net
peer
name,
something
address
set
to
Unix
domain,
socket
I,
don't
even
think,
we've
populated
for
Unix
the
main
circuit
I
think
we
are
looking
at
it
from
stuck
at
pure,
socket
instrumentation.
E
That
doesn't
need
to
know
about
logical
thing,
but
this
way
can
change.
I,
don't
see
a
problem
if
we
at
this
at
least
currently,
if
we
stick
with
just
server
address
for
Unix
domain
sockets.
E
A
Let's
come
back
to
the
proxy
example.
B
B
So
basically
the
client
would
then
so
it's
calling
the
example.com,
which
is
the
address
of
the
server
right.
B
E
B
E
B
Okay-
and
this
is
only
relevant
for
HTTP-
or
do
we
have
other
types
of
connections
that
where
we
would
have
this
scenario.
E
As
far
as
I
remember,
there
are
database
proxies,
so
we
should
in
theory,
have
just
in
general,
IP
proxies.
B
B
B
A
Oh
yeah,
we're
we're
having
that
discussion
internally,
so
kind
of
I,
think
Ludmila
and
my
goal
is
to
map
out.
If
we
want
to
align
with
ECS.
This
is
what
it
would
look
like
and
then
the
community
is
the
broader
Community
is
going
to
have
to
decide
whether
you
know
the
alignment
is
worth
the
breakage.
B
A
Oh
yeah,
no
problem
the
server
Port
server
IP,
so
basically
removing
that
from
these
three.
E
So
in
open
Telemetry
we
want
to
be
able
to
express
my
peer
address
right,
so
the
reverse
proxy.
What
I'm
connected
to.
B
E
Oh
I
think
this
is
the
not
necessarily
reverse
proxy
case
Okay.
So
but
then
there
is
no
yeah,
so
I
don't
think
it's
ever
possible
to
populate
this
one
for
http.
A
Okay
cool,
so
one
thing
that
I
didn't
understand
how
to
map
then
still
is
on
the
server
side.
So
we
have
right
server,
IP
server
port,
but
we
also
have
sort
of
so
this
is
the
logical
name
or
what
the
client
sent
to
us
or
what
we've
our
server
name,
but
then
at
the
stock,
well,
I
guess
the
yeah,
the
sock
host
port
would
be
potentially
there's
a
different
or
listening.
A
E
A
Yeah,
so
in
the
Java
world,
for
example,
right
we
have
local
port
and
local
address.
A
A
E
Yes,
which
could
be
Rewritten
along
the
way,
but
this
is
the
best
we
can
get.
B
Okay,
the
point
is
that
in
ects
we
have
a
dedicated
namespace
for
host
information,
so
if
it's
about
like
describing
the
host
itself,
so
maybe
this
is
what
is
missing.
Maybe
yeah.
A
I
mean
it's,
but
I
guess
I
would
worry
that
this,
like
open
Telemetry,
also
has
resource
semantic
conventions
that
describe
the
host
that
it's
running
on,
which
wouldn't
necessarily
be
the
same
as,
like
you
know
this
matched
virtual
host
idea.
E
Case
like,
if
everything
is
done
absolutely
wonderful,
you
probably
register
multiple
resources
right
and
for
multiple
hosts.
If
you
was
several
of
them
and
then
they
still
come
as
attributes
right.
So
it's
an
interesting
question,
then
foreign
Telemetry
study,
whether
this
primary
server
address,
should
even
be
a
spin
attribute.
E
A
A
A
B
B
Proxy,
so
the
client
is
talking
and
knowing
only
the
the
proxy
address
and
Port
right.
E
B
E
E
Not
not
to
even
forward
it's,
for
example,
for
Docker
right.
My
clients
thinks
it's
for
first
three,
but
it's
not
what
I
know.
E
B
A
So
that's
yeah,
that's
here
in
the
the
net
host
name,
can.
E
B
A
D
A
An
example
not
I
couldn't
find
an
example
where
the
net
post
IP
is
necessarily
different
than
the
sock
peer
address,
but
for
sure
the
the
net
host
port
and
the
net
sock
tier
port
are
generally
different
again
because
we
capture
the
ephemeral
Port
here.
E
Yeah,
that's
a
great
Point,
like
information
about
myself
is
not
quite
useful,
right,
I
think
what
we
can
do
here,
I
think.
All
of
this
sorry,
they're,
optional,
I,
think
post
peer
address
is
optional
and
or
names
one
of
them
is
optional,
but
my
point
is:
if
we
know
how
to
evolve
this
attributes
in
future
with
ECS,
what
would
be
the
right
place
for
them?
E
I
think
we
can
I,
don't
know
we
can
try
to
figure
out
if
this
optional
attributes
makes
any
make
any
sense
and
if
we
can
remove
them
from
now.
A
So
Alex
does
it
sounds
like
these
as
sort
of
these
ones,
probably.
A
Because
I'm
just
I'm
guessing
host
name
is
more
like.
Is
it
not
physical
but
like
the
underlying
post?
Yeah?
A
A
Net
sock
host
net-
okay-
we
talked
about
these-
these
are
just
the
client
right.
Okay,
so
I
think
we
covered
all
of
those
net
peer
name
net
report.
A
That,
okay,
this
was
what
we
talked
about
as
the
new
field
for
this
one
for
the
proxies
knit
oh,
but
we
were
calling
it
net.
A
A
D
A
The
normal
HTTP
case
and
I
can
put
out
put
up
another
comment
with
the
updated
mappings
I.
Don't
really
understand
the
Unix
socket
stuff
conceptually
so
do
you
think
we're
covered
there
with
the
net
with
the
server
address
I
think
is
what
the
proposal
was.
E
E
Interesting
anyway,
so
I
think
we
if
we
are
creating
server
stock
anyway,
we're
basically
creating
a
sock,
sub
name
space
right
and
if
address
is
not
enough,
which
it's
not,
but
we
can
add
stuff
there
later
once
we
have
a
need
to
instrument
something
with
Unix
Dimensions
kits,
which
I
think
didn't
come
yet.
E
My
only
question
here
is,
it
seems:
ECS
is
mostly
like
more
flat,
should
it
be
in
the
server
sock
or
should
it
be
just
a
separate,
sulk
namespace
if
it
has
a
separate
one,
we
still
need
to
separate
the
client
and
server
side.
So
probably
no
right.
B
So
in
in
the
Cs
there's
also
this
concept
of
reuse
of
namespaces,
so
you
can
basically
Nest
one
namespace
completely
under
another
one.
So
you
could,
if
you
would
Define
a
new
sock
namespace,
you
could
just
Nest
it
also
under
server
and
clock,
for
example,
okay,.
A
Oh,
but
this
would
it
would
still
end
up
being
this
name?
Yes,
exactly
got
it
got
it
not!
Okay,
oh
and
that's
what
you
were
saying
Ludmilla.
It
still
needs
both.
It's
not
like.
We
can
just
have
a
shared,
socket,
namespace.
A
E
This
one
but
I
mean
we
don't
have
a
really
any
need
for
the
soccer
port
and
we
don't
yet
have
a
need
for
a
client
soak
a
dress.
So
we
can
keep
it
in
mind
once
we
need
it
or
once
we
do
like
more
thorough
check
right.
Maybe
you
will
discover
we
need
it,
but
until
we
have
a
strong
reason
to
let's
just
stay
with
one,
that
would
be
my
choice.
A
And
Alex
worked
on
Google's
even
suggesting
I
noticed
there's
not
as
many
abbreviations
in
ECS
would
you
do.
You
have
any
recommendation
from
ECS
side
on
sock
versus
sockets.
E
Alex,
I
wonder:
did
anyone
ever
express
some
concern
about
the
lens
of
the
attribute
names
because
essentially,
at
least
in
case
of
Microsoft
power
observability,
then
their
people
pay
for
ingested
bytes
as
well,
and
we
are
inflating
the
this.
This
number
I
understand
it's
probably
G,
zip
10,
that's
probably
nothing
much,
but
did
you
hear
any
concerns
ever
about
it.
E
Are
lots
of
attributes
lens
sorry
their
lines
of
attribute
names.
B
No
actually
not
I,
think
yeah,
it's
maybe
negligible.
If
you
could
just
consider
how
how
many
fields
and
attributes
we
are
collecting
in,
like
in
In
Sum,
okay,.
A
Okay,
awesome,
so
I
will
put
I'll
put
out
a
new
comment
with
updated
mappings
and
oh
so,
let's
see
do
we
have
any
other
open
questions.
E
B
D
E
We
are
just
Uber
nerds,
we're
not
sure
if
you
like
it
so
like
I
mean
it
would
be
nice
to
have
some
place
to
put
protocol
name
too,
but
I'm
not
sure.
If
we
need
to
go
all
the
way
down
to
any
you
know
all
level
stuff.
We
also
have
Network
transport,
which
captures
it.
This
anyways.