►
From YouTube: 2021-01-13 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
A
A
A
A
B
I
was
actually
just
looking
over
this
versioning
and
stability
doc
pr
and
was
looking
at
some
of
your
comments
insightful.
As
always.
B
B
What
matters
is
that
the
guarantees
are
provided
for
not
how
exactly
they're
done,
and
so,
if
there's
a
strategy
in
java
for
being
able
to
seamlessly
transition
from
experimental
to
stable
that
allows
the
experimental
stuff
you
know
to
not
mess
up
the
stability
guarantees
the
other
stuff
and
not
force
the
users
of
the
experimental
stuff
to
have
a
big
breaking
change
by
design.
B
B
You
know
1.0
and
then
like
a
a
1.2
preview
that
has
all
the
metrics
work
on
it
and
they'll.
Just
they
just
want
to
commit
to
maintaining
two
separate
branches,
because
if
you
keep
the
codes
separate,
it
is
like
a
separate
enough
concept
that
you
know
that
form
of
just
porting
work
between
branches
is
like
theoretically,
not
that
hard.
As
long
as
you
actually
do,
keep
it
separate.
B
And
so
then
the
people
who
want
the
metrics
stuff
know
what
they're
getting
into
because
they're
explicitly
asking
for
the
preview
version,
which
I
think
means
in
nougat
or
whatever
their
package
manager
is
that
won't
be
like
the
default
version
that
you
get
so
that
all
sounded
pretty
cool.
So
these
are
like
three
strategies
now
that
have
popped
up
that
are
like
reasonable,
so
yeah.
C
B
Yeah,
there's
definitely
like
room
for
like
cleanup
and
saying
like
this
is
where
we
break
all
the
deprecated
crap
at
some
point
before
it
goes
stable
yeah.
It's
just
there's
like
the
strategy
of
like
moving
things
in
a
way
that
would
like
break
import
paths
seem
to
be
like
in
all
the
languages
I
talked
to.
B
That
was
like
the
first
solution,
everyone
thought
of,
and
I
was
like.
Oh
no,
I
know
there's
we
could
do
better
than
that,
and
now
we've
heard
a
bunch
of
bunch
of
patterns.
So
I'm
gonna
try
to
rewrite
this
just
to
identify
what
the
goals
are
and
then
give
a
couple
of
examples
of
like
how
you
could
how
you
could
achieve
those
goals.
B
Thing
yeah
bogdan:
we
were
just
going
over
this
versioning
and
stability
pr
and
looking
at
like
the
last
bits
that
were
open
and
one
of
the
main
threads
was
around
how
to
go
move
from
experimental
to
stable
and
we've
been
there's
like
a
strategy
laid
out
in
the
dock
that
explicitly
states.
You
should
build
it
in
a
separate
package
and
then
the
global
package
should
link
to
it,
but
people
are
pointing
out
in
a
couple
different
languages
that
they
could
achieve
the
same
goals
using
like
a
strategy
they'd
prefer.
B
E
One
of
the
things
that
you
mentioned,
and
you
need
to
point
out,
is
that
code
written
against
the
rc
if
no
breaking
changes
are
added,
for
whatever
reasons
I
mean
during
the
rc,
you
may
discover
that
the
the
planet
destroyed
itself
or
whatever
yeah.
Then
then
it
is
fine
to
break
things,
but
otherwise
you
should
not
break
things.
B
Right
like
when
we
hit
the
rc,
that's
that's
the
point
at
which
we're
trying
to
act
as
if
we
1.0
mm-hmm
right
or
we
act
as
if
you've
gotten
stable
and
see
see
if
you
can
keep
it
up.
B
It
it
does
happen
that
that's
the
point
at
which
you
receive
some
feedback.
That
makes
you
go
oh
yeah,
so
yeah
it
does
happen
but
yeah
the
point
is
you
think
it's
not
going
to
that's
why
you're
doing
nrc
yeah?
Okay?
I
think
people
get
that.
But
that's
that's
a
good
point.
We
should
make
it
explicit.
B
Tigran
was
wondering
about
this
is
a
total
bike
shed,
but
we
were
using
the
term
language
library
in
one
part
of
this
dock
or
one
part
of
the
spec,
and
then
in
this
document
I
was
using
the
term
open
telemetry
and
was
asked
to
make
it
more
specific.
So
I
started
calling
this
stuff
the
open,
telemetry
clients,
because
that
was
what
they
seemed
like
to
me.
We
pointed
out,
we
have
like
an
equivalent
term
called
language
library
elsewhere
in
the
doc.
C
B
B
That
stuff
we
were
calling
a
language
library
before
or
language
implementation,
and
here
I
started
calling
it
the
open,
telemetry
client,
because
it's
sort
of
the
client
architecture
in
our
topology,
but
do
people
have
a
preference
here.
C
C
C
B
B
B
We're
gonna
we're
gonna
ship.
This
thing,
yeah
yeah
did
you
have
anything
on
your
agenda.
B
E
E
E
For
for
the
batch,
for
example,
we
do
have
the
disrupter
version
yeah,
which
does
the
same
thing
but
uses
disrupter,
which
is
a
fat
dependency,
and
we
don't
want
to
have
that
in
the
sdk.
So
I'm
just
telling
reasons,
I'm
not
not
doing
anything
so
yeah
this.
E
These
are
the
two
reasons
that
I
know
of
the
other
thing
was,
but
this
is
arguably
for
me,
but
maybe
because
I'm
one
of
the
maintainer
and
author
of
this
it's
much
clearer
to
to
install
my
own
instance
there,
but
that
may
not
be
true
for
for
other
users.
Other
users
may
want
to
just
set
up
some
some
properties
and
be
done
with
this.
So
I
don't
know
yeah.
E
Yes,
so
so
you
can
do
the
filter
before
converting
to
spandex
or
after
and.
C
C
E
E
C
E
But
I
think
I
think
the
auto
configuration
or
the
configuration
mechanism
should
be
able
to
install
it,
even
though
it's
an
instance
we
there
is
a
separate
question
if
the
builder
pattern
or
auto
configuration
pattern,
whatever
pattern
you
use
to
create
a
big
provider
instance,
it
should
be
able
to,
or
should
be
aware
of,
the
existencing
of
the
batch
processor
and
install
it
in
the
right
place
for
for
the
like,
lazy
user,
not
lazy
user,
for
the
for
the
beginners
or
for
for
not
advanced
use
cases.
C
C
B
Is
there
any
any
further
work
we
need
to
do
around
aligning
configuration
across
languages
like
we
don't
currently
have
a.
We
have
a
concept
of
collectors
getting
configured
from,
say,
yaml
file.
We
do
have
like
some
concepts
of
environment
variables,
but
it
feels
like
we
don't
have
something
like
a
schema
configuration
schema
or
anything
partially.
That's
like
maybe
not
reasonable,
because
the
languages
are
too
different,
but
I
don't
know
there's
like
this.
B
This
happy
path,
which
I
think
is
what
you're
after,
where
like,
if
you're,
attaching
yourself
to
an
endpoint,
that's
like
otlp
compliant,
you
should
be
able
to
just
copy
paste,
a
single
line
and
then,
like
add
your,
you
know
ip
as
like
an
environment
variable
and
be
like
that's
it
yeah.
C
C
One
way
of
configuring
is
using
a
builder
or
something
and
usually
we'd
expect
this
to
be
sort
of
the
ideal
experience,
and
it
should
have
as
little
as
few
concepts
as
possible
that,
like
maybe,
if
we
have
explanations
of
these
sort
of
use
cases
that
could
be
helpful.
But
if
we
get
too
into
details
any
other
languages,
I'm
sure
you're
gonna
have
different
ways
of
actually
implementing
those.
B
I
think
I
need
to
like
do
another
round
of
setting
everything
up
and
seeing
how
it
how
it
currently
feels
yeah,
there's
also
the
reality
of
like
most
back
ends
require
some
amount
of
configuration
to
talk
to
that
back
end,
but
a
lot
of
that
I'm
noticing
can
get
bundled
up
like
for
light
step,
we're
able
to
bundle
up
our
configuration
it's
minimal,
but
likewise
for
like
jaeger,
prometheus
and
some
of
these
other
things
or
aws
and
splunk
and
everybody
else
like.
B
I
wonder,
is
there
a
way
we
could
find
some
convergence
there
so
that,
like
it's,
not
helpful
to
make
an
easy
default
if,
like
no
back
end
uses
that
default?
I
guess
that's
what
I'm
trying
to
say
so
is
there?
If
there
are
blockers
to
everyone
being
able
to
switch
to
vanilla
otlp,
we
should
figure
out
what
that
is,
and
maybe
fix
it
as
like.
B
A
as
like
a
a
group,
make
sure
that
we're
all
figuring
out
a
way
to
collect
this
data
in
such
a
way
that,
like
the
end
user,
is
able
to
like,
if
we
standardize
enough
on
our
end,
then
that
will
enable
them
to
to
standardize
more
on
their
end.
B
C
C
C
D
C
E
What,
if
it's,
if
it's
just
the
internal,
you
can
always
change
correct?
If,
if
you
don't
have
to
change
the
interface.
C
E
You
can
do
that.
The
only
problem
with
that
is
the
the
functionality
like
the
z
pages
functionality,
which
do
not
require
to
convert
them
to
spam
data,
always
so,
and
they
just
keep
keep
reference
to
the
span
and
convert
them
to
the
span
data.
Only
when
you
go
to
the
handler
that
serves
this,
that
will
will
be
penalized
by
by
doing
this.
E
E
D
C
An
exporter
like
use
the
export
interface
instead
of
the
spam
interface
for
that,
and
so
tracer
provider
runs
through
all
the
spam
processors
and
then
at
the
end
it
calls
to
spandate
and
passes
it
on
to
all
the
exporters.
One
of
them
could
be
a
batch
span.
Exporter.
E
Essentially,
essentially,
okay,
I
think
it's
not
too
late
for
that.
If
no,
no,
because
it's
not
too
much
of
a
change,
what
you
need
is
you
you?
E
You
always
have
the
simple
process,
simple,
what
we
call
simple
into
the
provider
yeah,
so
so
provider
exposes
to
two
ways
to
to
interact
with
the
spans
via
spam
processor,
which
you
have
on
start
on
end
or
via
this
exporter
chain,
and
then
you
build
the
the
the
patch
there
again
keep
in
mind
that
that
extra
work
to
convert
from
spam
to
spam
data
in
in
this
design
happens
on
critical
path.
E
Compare
with
right
now,
where
we
we
batch
spans,
so
no
extra
work
and
then
we
convert
asynchronously.
So.
A
E
I
still
believe
it's
not
impossible
to
do
it
even
after
g
with
some
deprecations
and
some
some
things
we
probably
is
not
too
much
to
to
to
change.
A
C
E
E
Oh,
we
also
discussed
about
the
the
service
name,
the
making
that
required
or
not,
and
I
think
I
think
the
final
decision
was
that.
E
I
don't,
I
don't
even
remember
the
final
decision
anyway,
the
final
decision
was
okay,
yes,
resource
will
expose
a
get
empty
or
get
default
and
get
default
will
have
the
default
values
for
for
service
for
a
couple
of
things
that
we
want
to
have
default
values
and
and
then
and
then
you
can
build
on
top
of
that,
you
can
add
more
things,
merge
more
resources
if
you
start
from
default
and
if,
if
there
is
anywhere
else
where
you
overwrite,
you
will
override
the
spanning
the
service
name.
E
Otherwise,
you
don't
so
so
resource,
essentially
a
standalone
package.
We
will
just
expose
a
get
default
instead
and
and
we
encourage
people
to
use
the
get
default
and
start
from
there
to
merge
other
resources.
On
top
of
that,
and
when
you
build
something
you
will
have
a
service
name
either
the
default.
If
you
did
not
overwrite
it
or
the
overwrite
value.
C
E
Okay,
it's
not
you
always
margin
to
default.
You
can.
You
can
start
from
empty
and
okay
yeah,
but
but
we
we
recommend
people
to
start
from
default.
D
D
B
Yeah,
well,
I'm
glad
to
get
the
naming
and
ib
id
bit
kind
of
resolved,
and
I
am
happy
that
we're
kind
of
pushing
people
towards
including
those
things,
because
I
think
related
to
that
earlier
comment
around
like
trying
to
figure
out
what
what
a
default
configuration
would
look
like
that
everyone
would
be
able
to
build
a
receiver
for
without
needing
their
own
specialized
configuration.
B
C
C
C
Never
mind
there
was
one.
There
was
still
the
discussion
with
the
duration.
C
E
But
that's
that's
a
separate
pr
correct
right!
That's
a
separate.
E
So
I
think
the
this
one,
the
split
variable
it's
approved
and
no
action-
I
mean
only
approves
in
the
past,
so
we
can
go
ahead
and
merge
this
one.
E
One,
let's
okay,
I
mean
I,
I
don't
have
a
good
explanation
right
now,.
E
C
C
E
Should
okay,
we
need
to
to
to
address
josh
josh's.