►
From YouTube: 2023-02-22 meeting
Description
Open Telemetry Meeting 1's Personal Meeting Room
B
B
Looking
for
I
think
aspects,
I
just
had
it
to
walk
through
all
the
existing
issues.
I,
just
probably
I,
may
have
missed
lots
of
them,
but
just
something
we
just
scratched.
My
eye
was
again
it's
basically
API
change
in
Trace
API
to
add
links
to
the
spine
after
the
span
is
created,
so
they
have
been
multiple
Alternatives,
which
has
been
discussed
and
the
PRS
has
been
released.
So
there
is
another
alternative
which
has
been
now
there's
a
PR
base
for
that
to
add
the
record
link
for
spam.
B
So,
instead
of
having
a
ad
link
method,
it's
something
called
a
record
link,
just
probably
just
to
being
consistent
with
the
naming
for
record
exceptions,
which
means
you
can
use
add
even
behind
the
scenes.
Okay,
okay,
so
this
can
internally
use
add
events
to
store
all
the
attributes,
but
yeah.
This
is
still
being
discussed
still
both
these
both
the
PRS
are
open.
So
both
are
the
API
breaking
changes.
B
So
I
think
this
is
something
which
probably
we
have
to
follow
up
more
closely
and
again,
I
see
that
somebody
has
again
the
questions.
Some
changes
on
this.
B
Okay
but
yeah,
so
this
is
this
is
being
something
which
probably
we
have
to
follow.
Nothing
from
that.
The
other
thing
is
rename
the
log
API
to
log
Bridge,
API,
so
yeah.
This
is
this
is
already
merged
now,
and
this
is
something
probably
we
may
want
to
do
any
changes
that,
let's
see
probably
they
if
there
is
any
change
required
in
the
API
I'm,
not
sure
but
yeah.
Probably
if
you
want
to
be
more
clear
on
the
API
enabling
you
may
have
to
change
our
class
name,
so
log
Bridge.
B
Something
is
required
here,
so
yeah
there's
something
probably
I
have
to
see
if
I
basically
raise
the
issue
for
this,
and
probably
let's
see
if
there's
any
change
required
for
us
in
the
API.
B
Yeah,
this
is
a
small
thing,
but
probably
the
this
ads
in
the
instrument
selection
idea,
it
adds
a
unit
attribute
unit
field.
Also,
so
just
need
something
this
in
the
code.
So
right
now
we
were
just
using
name
version
and
schema
URL,
but
I
think
it
also
be
able
to
actually
build
it
while
matching
the
while
matching
the
viewer
instrument.
So
as
part
of
view
configuration,
we
have
to
add
this
here.
B
B
B
Okay,
this
just
adds
sdk.md
and
api.md
200
specifications,
so
hopefully
I
think
this
will
not
give
any
changes
in
our
site.
A
B
B
B
Okay,
I
think
you
can
go
to
the
PRS.
Probably
we
have
small
number
of
PR
so.
B
D
B
A
B
Exactly
yeah,
all
only
all
the
cash
here
is
that
we
are
going
to
add
it
in
the
main
repo.
So
we
want
somebody
who
has
understanding
of
it
in
this
case,
probably
this
guy
to
really
maintain
it
for
some
time,
at
least
for
doing
it.
Sorry,
you
have
to
set
that
expectation
yeah.
You
will
set
that
expectation
with
him
as
as
none
of
us,
and
it
makes
sense
and
I,
don't
think
anybody
will
have
more
idea
about
one
two
three
should
be
somebody
to
maintain.
B
I
also
had
a
concern
of
using
open
tracing
repository
as
a
dependency,
but
I
think
I
did
review
the
pr
and
I
felt
that
removing
this
dependency
will
need
lots
of
code,
copying
lots
of
structures
and
classes
from
that.
From
this,
therefore,
to
either
it
will
need
lots
of
changes
or
lots
of
code
would
be
added
in
in
this
PR.
If
we
try
to
remove
this
dependency,
so
I
difficult
to
get
rid
of
this
dependency
in
the
pr.
A
A
B
A
And
also
the
more
so
it's
good
that
we
have
this,
because
we
need
that
to
have
that
perfect
spec
for
12,
open
tracing
shim,
but
also
at
the
same
time,.
B
A
B
I
think
that
that
definitely
makes
sense
I
think
we
probably
with
our
limited
understanding
of
open
tracing
the
code,
looked
cleaner
without
any
any
without
affecting
any
of
the
our
existing
code
base.
So
probably
I
think
I
can
spend
some
time
today,
or
at
least
this
week
for
sure,
and
let's,
let's
have
another
review
on
this.
If
two
of
us
yeah,
I,
think
Mark,
you
have
already
approved
it.
If
I
think
one
of
us
approve
it
I
think
it
should
be
good.
B
B
Okay,
yeah
I
think
I
put
this
Matrix
performance.
Improvement
I
have
been
working
past
a
few
days
or
measuring
the
performance
of
our
Matrix
code
and
probably
see
the
areas
where
we
can
do
some
improvements.
I
did
find
find
some
areas.
At
least
we'd
have
raised
a
PR
for
that,
so
probably
I
just
was
having
a
benchmark
of
how
much
time
it
takes
to
process
around
1
million
measurements.
That
was
just
a
number
which
I
took
it
probably
and
then
started
seeing
in
it
homework.
B
And
that's
has
I
think
were
good
good
number.
The
number
is
really
promising,
I
think
at
least
from
what
I
could
see
here
we
could
achieve
from
Maria.
It
was
taking
around
around
3.7
seconds
to
process
one
million
measurements
with
three
dimensional,
and
each
Dimension
can
have
10
different
combinations,
so
I
think
around
one
around
one
thousand
combinations
of
different
dimensions
and
having
measurements
of
one
million
measurements,
assuming
that
there
is
no
collection
happening
in
between,
because
collection
is
a
bit
heavy
process.
B
So
assuming
there
is
no
connection
happening,
how
much
time
it
takes
to
process
one
million
measurements,
so
with
this
PR
at
least
I
am
able
to
achieve
from
3.7
seconds
to
1.9
seconds
again.
This
is
assuming
that
there
is
only
one
thread
which
is
sending
all
these
measurements,
and
that
was
the
the
CPU
configuration
which
I
was
using
for
this.
B
This
is
a
different
I
mean,
as
I
said,
that
there
is
no
collection
happening
as
of
now
in
this,
so
I
did
use
ostream
exporter,
but
our
periodic
using
periodic
reader,
but
but
the
collection
time
was
quite
large
so
to
ensure
that
within
this
three
four
seconds
there
was
no
collection
happening.
So
this
is
just
the
performance
of
SDK,
not
the
performance
of
collector
or
exporter.
Okay,
yeah.
B
B
I
still
have,
there
are
still
some
issues
here,
because
this
was
done
using
one
you
as
I
said:
there's
only
one
side
was
there,
which
were
which
was
sending
all
the
measurements,
but
when
I,
when
I
try
to
scale
it.
B
So
this
this,
this
configuration
was
having
around
20
processes,
but
when
I
try
to
scale
to
multiple
threads
I
could
not
get
a
linear
performance
Improvement
and
which
makes
sense
because
we
use
some
locks
and
there
is
a
common
internal
buffer
which,
where
we
have
to
store
all
them,
we
will
be
aggregate,
install
all
the
measurements
and
that
buffer
is
under
the
lock.
So
I
think
the
Improvement
is
not
to
the
extent
where
I
was
I
mean
where
it
should
be,
at
least
as
we
as
we
scale
the
number
of
threads.
B
So
probably
there
is
just
scope
of
improvement.
I
will
be
checking
that
also
separately
and
I.
Think
we
also
discussed
earlier
if
we
can
have
some
log
free
data
structures
like
maps
and
other
Maps,
which
we
can
use
in
these
scenarios
and
I
I
didn't
find
earlier
when
we
were
implementing
this
any
any
easy
way
of
using
a
locked
free
data
structure
to
store
our
measurements
and
aggregated
data.
But
yeah,
probably
I'll
have
more
look
into
that
that,
because
that
can
help
us
scale
scale
the
performance
as
we
increase
the
number
of
threads.
B
So
still
some
scope
of
my
Improvement
yeah
just
just
wanted
to
call
out.
B
Okay,
yeah,
apart
from
that
yeah
in
miscellaneous
item,
I,
think
we'll
go
with
issues
starting
afterwards
when
people
through
one
of
them
Avi
check
I,
think
that's
a
very
good
point:
marketing
tool
to
detect
EDH
changes
in
CI
instead
of
relying
on
the
code
removed-
and
this
has
been
discussed
multiple
times
early
very
early
in
I-
mean
during
the
trace,
Tracer
implementation
and
Max
did
some
investigation
on
the
API
compliance.
B
Checker
I
think
he
was
kind
of
satisfied
with
this
kind
of
tool
and
I
think
probably
he
is
I
think
this
compares
the
old
API.
It
takes
the
dump
of
the
old
API
in
a
new
API
and
then
do
it
from
code
environment
comparison
to
find
out,
but
yeah
I
mean
it's.
Definitely
it's
a
very
good
idea.
If
you
can
have
any
tool,
if
someone
can
definitely
investigate
to
have
any
tool
which
can
be
part
of
CI,
so
that
in
a
non-avi
breaking
changes
can
be
caught
much
upfront
we
can,
we
can
have
a.
A
Yeah
just
looking
at
the
recent
changes,
so
there
is
a
Visa
training
in
respect
where
people
want
to
add
links
to
respond,
and
that
would
be
a
change
in
the
API
and
therefore
in
the
Abi.
There
is
another
change
just
today
that
we
saw
that
in
views
for
metrics.
We
want
to
add
units
and.
E
A
Time
there
is
a
change
like
that:
I
mean
if
we
add
a
new
parameter
to
an
API.
It's
a
breaking
change,
even
if
we
make
that
parameter
optional,
it
may
compile
of
Revolt
code,
but
it's
still
a
binary
change,
because
some
API
will
expect
three
parameters,
no
expect
five
and
it's
it
will
crash
and
every
time
we
add,
even
if
we
add
a
new
method
for
a
new
feature.
If
it's
a
new
virtual
method,
it
also
breaks
the
virtual
table
of
a
class.
So
it's
a
breaking
change
and
so
forth
and
so
forth.
B
Yeah
so
I
mean
this
will
give
a
brief
of
this
brief
history.
Here
by
default,
our
API
was
supposed
to
be
Avi
compliant.
It
is
supposed
to
be
a
very
complied
and
if
there
are
any
breaking
changes,
we'll
just
increment
the
ABI
version
of
the
of
the
API
and
the
the
intention
behind
incrementing,
the
Avi
version
was
to
catch
any
Avi
breaking
issues
during
link
time
and
those
should
not
go
all
the
ahead.
I
mean
and
we
should
not
have
any
weird
memory
access
issues.
B
A
D
A
The
new
code
will
not
so
they
will.
This
will
use
different
symbols
in
the
binary,
so
there
will
be
no
mid
snatch
between
vocal
and
the
new
code.
So
this
is
one
thing.
However,
I
looked
at
the
specification
themselves
and
in
respect
it
says
that
once
an
API
has
been
exposed,
it
basically
should
be
exposed
implemented
forever,
so
that,
let's
say
we,
we
do
some
change
when
we
expose
ABI
number
two.
E
B
A
It's
it's
likely
to
be
to
be
difficult.
Yes,.
B
You
know:
that's
that's
that
that
was
the
only
cat,
so
I
think
we
just
wanted
to
keep
it
simple
so
just
showing
so
the
only
to
keep
it
simple.
We
just
decided
to
ensure
that
we
should
not.
We
should
be
able
to
catch
all
those
stuffs
and
not
not
make
and
not
really
mentally,
aware
compatibility
if
you
want
to
if
we
want
to
do
any
changes,
but.
B
I
think
I
it's
it's
we
can
discuss
like
if
you
want
to
really
maintain
last,
probably
the
latest
two
versions
and
all
those
things
not
not
maintained
all
the
way
or
anything,
I
think
that
it's
totally
you
can.
E
A
Yeah,
at
least
we
need
to
clarify,
but,
and
especially
so,
the
trace
API
made
change.
We
have
seen
that
the
Matrix
API
may
change
for
units
and
you
did
some
performance
Improvement,
but
also
this
week,
I
I
stamped
on
a
performance
issue
in
the
info
metrics
asynchronous
API.
There
are
things.
B
B
A
B
B
A
Synchronous
counter
to
a
meter
to
today
the
the
create
asynchronous
counter:
I
returns,
a
variant
and
with
a
shared
pointer
and
things
like
that
and
bloodbot
can
be,
can
be
improved,
I
think,
but.
B
Yeah,
can
you
please
raise
the
issue
for
this?
Probably
yes,
any
such
issue
you're.
Finding,
let's,
let's
at
least
have
something
for
tracking
and
I,
would
be
definitely
interested
to
see
sure
any
such
issues
which
yeah.
B
Yeah,
so
so,
for
this,
at
least
for
any
ABI
breaking
thing
as
I
was
saying,
I
think
it's.
We
will
scope
it
for
the
open,
Telemetry
API
breaking
changes.
Yes,
we
want.
So
if
there's
any
change
in
the
SDK,
we
won't
maintain
all
the
versions.
In
that
scenario,
if
we
have,
if
you
plan
to
maintain
the
origins,
yeah.
A
B
A
Is
for
what
I
found
in
respect?
This
is
what
respect
says.
Actually
we
are
to
to
do
to
to
provide
compatibility
with
the
API
because
of
instrumented
application
may
be
out
there,
but
for
the
SDK.
B
B
Okay,
probably
I
think,
let's,
let's
I,
think
that
I
don't
think
we
can
decide
anything
on
this
in
this
meeting,
but
yeah
I
I.
Let's
continue
discussing
at
least
maintaining
multiple
version
that
that's
something
probably
I
think
we
have
to
think
more
before
really
deciding
on
that
sure.
Yeah.
B
A
Well,
so
I'm
not
sure
what
the
proper
media
is
for
that,
if
it's
a
discussion
or
a
PR
with
the
we
will
document
to
describe
things
I'm,
not
sure.
Yet
as
a
side
note,
I
hope
you
can
hear
me.
Okay,
I'm
I'm
at
sick
this
week
and
I
will
take
some
vacation.
B
B
B
E
B
Okay
and
that's
all
In
terms.
B
B
B
B
B
B
I
mean
yeah,
so
I
mean
the
only
last
last
week,
I
think
I
had
was
able
to
reproduce
the
issues.
That
was
only
things
I
did
to
the
it's
a
genuine
issue,
and
it's
something
to
do
with
with
the
Alpine
providing
its
own
grpc
packages
looks
like
it's
something
something
to
do
with
that,
but
yes,
it's
the
prodigible.
So
it's
a
validation
and
we
have
to
look
into
that.
B
B
B
Let
me
have
to
follow
it
up.
Sorry
I
think
I
have
to
follow
it
up
again.
B
B
This,
the
old
one,
can
we
know:
where
are
you
on
this.
B
B
B
B
We
can
do
anything
for
etw
exported
if
you
can
move
it
to
contract,
sure
it's
all
header,
so
it
may
not
be
very
difficult
to
do
it,
but
let
me
check
if,
if
something
which
we
can
achieve
it
yeah.
B
B
B
So
if
you
see
your
exporters,
we
have
these
two
are
old,
something
which
I
have
been
maintaining
it
and
I
have
to
add
it
and
I
know.
Tom
has
been
asking
me
to
to
add
a
CI
for
these,
both
of
these
and
I.
Think
somehow
I
have
to
find
time
to
add
and
give
me
case.
I,
don't
think
it's
this.
B
A
E
D
B
I
already
already
discussed
it,
you
know
we
know
what
what
exactly
it
is.
Hopefully.
B
Yeah,
this
is
draft,
so
I
think
I'm,
not
looking
into
that
and
okay
continuing
on
this.
There
is
another
thing
which
you
told
about
TLD,
which
also
I
think
supposed
to
be
moved.
As
a
third
party,
Library
I
mean
the
with
TLD
I
mean
I,
thought
I
mean
I
went
three
slopping
Dynamics.
B
B
B
Yes,
there
is
a.
There
is
a
therefore
I
think
it's.
This
is
open
source
now,
so
I
think
it
should
be
easy
to
use
it
as
a
module,
but
I
think
both
can
be
done
together.
Moving
this
to
contrib
and
then
using
this
as
a
some
model,
let
me
see
you
can
achieve
it
for
elastic
search,
I'm,
not
sure
if
we
should
be
doing
it.
This
was
done
by
Amazon
and
I.
B
Don't
know,
I
mean
this
chart
of
the
project
and
I,
don't
know
what
to
do
if
you
can
just
remove
it,
unless
somebody
really
wants
it,
the
otherwise
we
can
just
probably
kind
of
if
we
create
it
or
just
remove
it
as
it's
in
one
standard
stuff.
You
know
in
our
main
report.
B
So
let
me
let
me
let
me
check
with
the
Amazon
team.
B
I
forgot
the
name
of
that
person,
but
I
think
I
can
check
with
them
what
what
do?
What
they
want
to
do
with
this,
as
we
cannot
keep
it
in
the
menu.
So
in
case
they
are
not
interested
to
maintain
it
or
anything.
You
can
just
drop
it
because
just
remove
it
from
The
Middle
and
that's
all
you
know,
moving
to
quantitative
I,
don't
know
how
can
we
do
achieve
it?
How
can
we
keep
the
build,
get
everything
for
that
may
not
be
very
straightforward.
B
B
B
B
Okay
and
then
the
last
one,
this
video.