►
From YouTube: 2022-02-24 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
Hey
everybody,
gemma
is
not
with
us
today
and
yeah.
So
there's
we
are.
We
have
only
one
issue
to
talk
about,
at
least
in
the
agenda
items,
so
we
can
either
discuss
that
or
or
disclose
something
else.
If
you
want
I'm
sorry
for
the
noise,
it
just
happens
that
my
neighbor
is
doing
renovation
in
my
flat,
so
it
means
I
have
to
work
from
a
coffee
shop.
Sorry
for
the
noise.
B
Okay,
hello,
yes,
so
we've
got
just
one
issue
as
suggested
by
by
josh:
I'm
not
sure
if,
if
you
guys
had
a
chance
to
to
have
a
look
at
it,
so
the
the
thing
is
well.
Maybe
I
should
share.
B
B
Does
it
does
it
work?
I
am
a
unable
to
share
for
whatever
reason.
B
B
B
Yeah,
so
I
understand
the
idea
here
is
that
we
have
this
probabilistic
sampling
processor,
which
really
means
that
we
sample
after
the
fact
after
the
collecting
the
the
trace
or
maybe
incomplete
trace.
So
this
might
be
happening
in
the
collector.
Perhaps,
but
it's
not
explicitly
stated
anyway
that
then
there
is
a
need
to
do
this
tail
based
sampling,
at
least
people
feel
that
there
is
a
need
for
that,
and
I
I
don't
disagree,
but
how
to
do
it
in
such
a
way
that
we
save
all
the
good
properties
of
our
unbiased
and
sampling
is.
B
B
B
Which
got
a
response
from
ottmar?
Unfortunately,
I
don't
see
lot
more
here,
but
down
there.
There
is
a
response
from
ottmer
that
he
wouldn't
really
like
to
touch
the
p-value.
If,
if
this
is
inconsistent,
sampling
really
being
applied
to
which
I
responded
that
perhaps
we
could
explicitly
show
the
adjusted
count
by
another
value
which
would
be
used
if
we
use
this
probabilistic
sampling,
processor.
B
Of
course,
this
does
not
cover
all
the
cases
we
we
don't
know
how
many
steps
processing
steps
a
trace
will
take.
There
is
no
limit
on
that,
so
every
step
needs
to
understand
what
it
receives
and-
and
my
concern
is
that
we
should.
We
should
have
well
defined
semantics
of
these
fields
like
r,
p
and
c
and
possibly
t
which
is
in
mentioned
in
the
latest
message
from
josh,
which,
which
is
in
the
slack
channel,
actually.
A
Yeah,
that's
something
interesting.
First
of
all,
I
think
that
one
of
the
issues
that
I
saw
is
that
the
way
you
know
the
random
generation
is
different,
so
maybe
we
need
to
change
the
that
processor
first
to
do
you
know
what
we
are
doing
in
the
new
pr.
Sorry
in
the
new
section.
The
second
thing
is
yeah.
What
you
talk
about,
I
didn't
know
about
the
part
only
about
the
c1
that
makes
sense
to
me.
A
B
B
A
I
guess
that
one
question
that
I
have
is
whether,
as
I
said
before,
I
think
we
should
prototype
this.
I
have
mentioned
my
detail.
We
could
probably
have
somebody
in
my
team
maybe
spend
some
cycles
there.
You
know
like
prototyping
this,
the
sea
approach.
At
least
you
know
in
this
case,
yes,
yeah,
okay,
sounds
okay
sounds
like
yeah.
B
Yeah,
that
would
be
great
if
we
could
have
some
on
hands
experience
here.
B
There
is
already
a
prototype,
at
least
for
for
for
for
java
written
by
ottmar,
and
I
heard
there
is
also
a
go
version
which
I
haven't
seen
so
maybe
future
work
can
be
leveraged.
On
top
of
that.
A
Yeah,
well,
actually
that
they
are
like
different.
Although
related
approaches,
the
prototypes
in
java
go
are
about
putting
the
the
r
and
the
and
the
p
values
from
the
tracers
or
the
client
side.
You
know
so,
once
you
are
creating
new
spans,
you
actually
are
setting
that
and
then
you
can
see
that
on
the
back
end.
B
B
Right
what
I
mean
here
is
that
this
is
this
is
at
least
a
way
to
generate,
hopefully
correct,
r
and
p
values.
So
so,
when
dealing
with
a
probabilistic
sampling
processor,
you
at
least
will
have
a
correct
input
stream.
A
Yeah,
that's
the
one.
I
think
that.
So
honestly,
I
don't
know
more
details,
but
I
think
that
one
of
the
initial
ideas
that
we
have
is
that
we
support
this
kind
of
tail
sampling
when
you
don't
have
any
p
or
r
air
value
in
the
spans.
You
know
like
nobody's
using
those
new
prototypes
in
java
and
go
so,
and
you
just
want
to
try
this
new
feature
the
adjusting
count
concept
from
the
collector
perspective.
A
A
Okay,
in
any
case
yeah,
I
think
that
yeah
well,
of
course,
as
I
said
before,
we
are
trying
to
get
some
cycles
from
my
team,
somebody
that
could
work
on
a
prototype
on
the
collector's
side,
probably
starting
without
reducing
p,
just
trying
to
set
the
p
value.
If
there's
no
p
value
existing.
Something
like
that,
just
as
a
proof
of
concept
and
then,
of
course,
we
would
probably
be
showcasing
such
pr.
A
The
only
thing,
the
only
question
that
I
have
for
the
moment
is
whether
we
should
change
the
the
randomness
there,
because
the
collector
this
this
existing
processor
uses
a
very
specific
randomness
generation.
It
uses
hashing
based
only
on
the
ids.
If
I
remember
correctly,
it
raised
a
dua
best
effort
and,
of
course,
in
the
case
of
this
new
section,
the
r
value
generated
differently.
A
Okay,
I
guess
that's
all
from
my
side.
I
will
try
to
poke
josh
once
he's
back
on
holidays,
and
I
will
try
to
I
said,
as
I
said
before,
somebody
my
team
to
have
a
prototype
on
the
collector's
side,
just
to
start
playing
with
what
could
be
done
there
but
yeah.
A
A
Okay,
I
think
we
are
cool,
then
thank
you
so
much
for
coming.
It
was
a
very
short
call,
but
at
least
we
know
that
we
should
start
working
on
a
prototype,
for
this
see
how
it
goes.
Thank
you
so
much
and
I
will
keep
you
posted.
Thank
you
guys.